Machine learning-based infection diagnostic and prognostic models in post-acute care settings: a systematic review

Zidu Xu; Danielle Scharp; Mollie Hobensack; Jiancheng Ye; Jungang Zou; Sirui Ding; Jingjing Shang; Maxim Topaz

doi:10.1093/jamia/ocae278

. 2024 Nov 12;32(1):241–252. doi: 10.1093/jamia/ocae278

Machine learning-based infection diagnostic and prognostic models in post-acute care settings: a systematic review

Zidu Xu ^1,^✉, Danielle Scharp ², Mollie Hobensack ³, Jiancheng Ye ⁴, Jungang Zou ⁵, Sirui Ding ⁶, Jingjing Shang ⁷, Maxim Topaz ^8,^9,¹⁰

PMCID: PMC11648729 PMID: 39530740

Abstract

Objectives

This study aims to (1) review machine learning (ML)-based models for early infection diagnostic and prognosis prediction in post-acute care (PAC) settings, (2) identify key risk predictors influencing infection-related outcomes, and (3) examine the quality and limitations of these models.

Materials and Methods

PubMed, Web of Science, Scopus, IEEE Xplore, CINAHL, and ACM digital library were searched in February 2024. Eligible studies leveraged PAC data to develop and evaluate ML models for infection-related risks. Data extraction followed the CHARMS checklist. Quality appraisal followed the PROBAST tool. Data synthesis was guided by the socio-ecological conceptual framework.

Results

Thirteen studies were included, mainly focusing on respiratory infections and nursing homes. Most used regression models with structured electronic health record data. Since 2020, there has been a shift toward advanced ML algorithms and multimodal data, biosensors, and clinical notes being significant sources of unstructured data. Despite these advances, there is insufficient evidence to support performance improvements over traditional models. Individual-level risk predictors, like impaired cognition, declined function, and tachycardia, were commonly used, while contextual-level predictors were barely utilized, consequently limiting model fairness. Major sources of bias included lack of external validation, inadequate model calibration, and insufficient consideration of data complexity.

Discussion and Conclusion

Despite the growth of advanced modeling approaches in infection-related models in PAC settings, evidence supporting their superiority remains limited. Future research should leverage a socio-ecological lens for predictor selection and model construction, exploring optimal data modalities and ML model usage in PAC, while ensuring rigorous methodologies and fairness considerations.

Keywords: prediction models, infection, post-acute care, machine learning, electronic health record

Introduction

Post-acute care (PAC) is an essential component of the US healthcare system, designed to support recovery and maintain the health of individuals following hospital discharge. PAC settings include long-term acute care (LTC) facilities, inpatient rehabilitation facilities, skilled nursing facilities/nursing homes (NHs), and home health care (HHC).¹^,² Collectively, these facilities manage the care of over 40% of Medicare beneficiaries annually and serve more than 5 million patients annually.³^,⁴ Approximately 85% of PAC patients are over 65, typically experiencing age-related physiological changes such as weakened immune systems, reduced mucociliary clearance, and slower wound healing.^5–7 These factors heighten the risk of infections, especially among patients with multiple chronic conditions and prior surgeries.⁴ These infections frequently manifest with atypical symptoms, often masked by other signs of deterioration in older adults, such as confusion and functional impairment,⁸^,⁹ leading to delayed diagnosis and treatment, causing severe complications, unplanned hospitalizations, emergency department visits, and increased mortality.¹⁰^,¹¹ Proactively identifying patients at high-risk of infections and poor prognosis is crucial for designing, implementing, and maintaining effective infection prevention and control programs to reduce adverse outcomes.¹²^,¹³

Prediction modeling has become an efficient strategy for clinical decision-making.¹⁴^,¹⁵ With advancements in health data digitalization and analytical approaches, machine learning (ML) methods are increasingly used to enhance infection-related risk prediction.⁸^,¹⁶^,¹⁷ ML, an umbrella term for a various set of statistical and computational algorithms, excels in automatically learning patterns and correlations between input features and the likelihood of specific clinical outcomes from historical electronic health record (EHR) data.¹⁸ This automation facilitates the prediction of outcomes from new, unseen data.

Recent studies have leveraged ML-based diagnostic and prognostic models to estimate infection-related outcomes, aiming to prompt individualized interventions for infection prevention and control. Diagnostic models assess the likelihood of an infection onset (ie, infection early detection), while prognostic models predict the probability of adverse outcomes such as mortality and infection exacerbation in those already diagnosed.¹⁹^,²⁰ These approaches are applied across various healthcare settings, including scattered studies focusing on PAC.¹⁷^,^21–28 Additionally, despite rapid evolution of ML techniques has shown increased capability in analyzing various data modalities to enhance infection-related risk predictions in acute care,²¹^,²²^,²⁹ their application status and impact within PAC settings remain unclear. Currently, there is a lack of systematic reviews on infection-related risk prediction models in PAC, underscoring the need for a comprehensive analysis of data types, predictors, modeling approaches and evaluation measures used, as well as implementation challenges in specific clinical scenarios. This will help identify knowledge gaps and guide further research to advance ML applications for infections in PAC.

Research has demonstrated that infection-related health outcomes in PAC settings are influenced by a hierarchy of interacting risk factors across multiple levels (Figure 1): individual patient characteristics (e.g., socio-demographics, comorbidities), interpersonal factors (e.g., housing conditions, informal caregiver availability),³⁰^,³¹ institutional characteristics (e.g., staff-patient ratios, specialist knowledge within a PAC facility),³²^,³³ community factors (e.g., proximity of health services, location rurality),³⁴^,³⁵ and broader societal influences (e.g., social norms, federal infection control policies).³⁶^,³⁷ To this end, the socio-ecological conceptual framework provides a holistic perspective to elucidate how multi-level factors contribute to infection-related outcomes in PAC settings.³⁸^,³⁹ By situating individuals within their specific PAC scenarios and adjusting for contextual attributes, this conceptual framework facilitates the creation of comprehensive clinical profiles, informing infection-related risk modeling, guiding clinical interventions, and minimizing avoidable disparities.³⁹^,⁴⁰

A socio-ecological framework illustrating multi-level factors influencing infection-related health outcomes in post-acute care settings, with levels from innermost to outermost: Individual, Interpersonal, Institutional, Community, and Societal. — The socio-ecological conceptual framework of health. This model (adapted from Bronfenbrenner et al.³⁹) can be used to summarize and understand the multi-level factors that influence infection-related health outcomes in PAC settings.

Objectives

This study aimed to (1) systematically review ML-based diagnostic and prognosis models for infections in PAC settings, (2) summarize key infection-related risk predictors using the socio-ecological conceptual framework, and (3) critically evaluate the quality (e.g., performance, clinical utility, fairness considerations) and limitations of these models. Findings will guide the development of rigorous and effective infection risk models in real-world PAC settings, facilitating the future integration of ML models into infection management practices.

Methods

Registration

This systematic review followed the guidelines outlined by the Preferred Reporting Items for Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines (Supplementary Appendix 1).⁴¹ The protocol was registered in the Open Science Framework (osf.io/chyej).

Eligibility criteria

This study selection aimed to identify research providing significant insights into the development, evaluation, and application of ML models for infections in PAC settings. We included primary prediction modeling studies that developed individualized diagnostic or prognostic for infection using data from PAC settings and provided detailed information on model development and evaluation. To ensure compliance with essential ML modeling stages—specifically, model development and either internal or external validation¹⁹^,²⁰^,⁴²^,⁴³—we excluded studies that identified predictors without performing any form of validation. Additionally, studies that solely evaluated existing models without updating predictors or algorithms were excluded, as they did not conform to the standardized evaluation criteria.²⁰ See Supplementary Appendix 2 for the full criteria.

Information sources and search strategy

The search strategy (Supplementary Appendix 2) was developed by the author (ZX) and refined in consultation with the co-authors and an Informationist from Columbia University August Health Sciences Library. Six databases (PubMed, Web of Science, Scopus, IEEE Xplore, CINAHL, and the ACM digital library) were searched through February 29, 2024, limited to English-language articles with no publication date restrictions.

Study selection and data extraction

Articles were de-duplicated using the Bramer method.⁴⁴ Each remaining title and abstract was screened, followed by full-text screening, with documentation of the reasons for exclusion. Each potentially eligible article was reviewed by at least two reviewers (ZX, DS, MH, or JY), and discrepancies were resolved through team meetings. A data extraction template was developed based on the Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies (CHARMS).²⁰ Two reviewers (ZX, DS, MH, and JZ) performed extraction independently, with final reviews by MT and JS to resolve discrepancies.

The data extraction template included fields for basic study characteristics, types of infections, PAC settings examined, ML models used, and primary outcomes. Additionally, we recorded the risk predictors included in the finalized models, along with their data types and sources, the best prediction performances reported, and other development and evaluation details.

Quality appraisal

This review utilized the Prediction Model Risk of Bias Assessment Tool (PROBAST)⁴⁵ to assess the risk of bias and applicability, which affects a model’s internal validity and trustworthiness.¹⁹^,⁴⁶ PROBAST uses 20 signaling questions across four domains (ie, participants, predictors, outcome, and analysis) to assign risk of bias ratings of low, high, or unclear.¹⁹ Two independent reviewers (ZX, DS, MH, and JZ) conducted independent assessments and then reviewed by MT and JS.

Data synthesis and conceptual framework

We used narrative synthesis, tabulation, and visualization to aggregate findings from data extraction process, identifying themes related to key research questions (Figure 2). Initially, we analyzed basic study characteristics and ML model applications. Next, we summarized the data types, data sources, and risk predictors utilized across studies. Then, we explored ML modeling approaches, validation strategies, and evaluation measures employed, including model performance metrics (e.g., area under the curve (AUC), sensitivity/recall, precision, and calibration), clinical utility, model interpretability, and fairness considerations. Due to significant inter-study heterogeneity, we reported the performance metrics of individual studies and provided a descriptive analysis to accommodate the diverse methodologies and outcomes.

A diagram describing the study selection process and research questions. On the left is a PRISMA 2020 flow diagram illustrating study identification, screening, and selection. On the right is a list of the research questions posed. — Study selection process and research questions. A. Flow diagram illustrating the PRISMA 2020 approach for the identification, screening, and selection of studies. B. Research questions posed.

To systematically categorize the identified risk predictors that cause and exacerbate infections among PAC patients, we applied the socio-ecological conceptual framework.³⁹ This conceptual framework recognizes that infection-related health outcomes are influenced by multiple, interacting levels of factors: individual, interpersonal, institutional, community, and societal.³⁸ By mapping the risk predictors onto these levels, the socio-ecological conceptual framework provides a structured and holistic approach to summarize the risk predictors used in infection diagnostic and prognostic models within PAC settings. This socio-ecological lens also guides the evaluation of model fairness by ensuring that Social Determinants of Health (SDoH) are thoroughly considered.

Results

Study selection

Figure 2 illustrates the article screening and selection process. We identified 3690 records through the selected databases, reducing to 3115 after deduplication. After title and abstract screening, 43 articles underwent full-text review. Hand search did not add additional studies. Finally, 13 studies met our eligibility criteria and were included in the review.

Study characteristics

Table 1 summarizes the characteristics of all included studies, which were published between 2000 and 2022 and used data from 1995 to 2020 (Supplementary Appendix 3). Studies predominantly focused on NHs (62%),^47–54 followed by LTC facilities (23%),³⁴^,³⁵^,⁵⁵ and HHC agencies (15%).⁵⁶^,⁵⁷ Most studies (62%) targeted respiratory tract infections.³⁴^,³⁵^,⁴⁷^,^50–54 The remaining studies addressed urinary tract infections,⁵⁵ wound infections,⁵⁷ or various infections together.⁴⁸^,⁴⁹^,⁵⁶ Overall, age, sex, and recruitment location were consistently reported (>60%). Other individual-level (e.g., race/ethnicity and socioeconomic status) and contextual-level SDoH (e.g., living arrangements and neighborhood deprivation) were poorly (<30%) reported (Supplementary Appendix 3). Most models developed since 2020 are diagnostic,³⁴^,^47–49^,^55–57 to predict infection onsets, whereas earlier were predominantly prognostic, targeting mortality risk post infections.⁵⁰^,⁵¹^,⁵³^,⁵⁴

Table 1.

Study summary.

Study	Settings	Infection types	ML applications
Lee 2022³⁴	LTC	RTI	1) Identify resident and community characteristics predictive of SARS-Cov-2 infection, and 2) determine best performed XGBoost algorithm; 3) risk stratification,
Ramazi 2022⁵⁵	LTC	UTI	1) Use RTLS and deep learning to classify motor behaviors; 2) predict acute events (falls, delirium, UTIs) in LTC residents with dementia.
Tan 2022⁴⁷	NH	RTI	Develop and validate a nomogram to predict the risk of NHAP.
Garcés 2021⁴⁹	NH	RTI, UTI, SSTI	Develop a system to classify acute RTI, UTI, or SSTI risk.
Lee 2021³⁵	LTC	RTI	1) Determine predictors of 30-day mortality after SARS-CoV-2 using ML; 2) develop a risk prediction model; 3) risk stratification.
Song 2021⁵⁷	HHC	SSTI	1) Identify risk factors for wound infections in HHC patients; 2) build wound infection predictive models.
Panagiotou 2021⁵²	NH	RTI	1) Identify 30-day all-cause mortality risk factors after COVID-19; 2) build mortality prediction model; 3) risk stratification.
Shang 2020⁵⁶	HHC	RTI, UTI, SSTI, IV catheter-related infection	1) Develop and test a patient risk model for infection-related hospitalizations or ED use regardless of its type; 2) risk stratification.
Baldominos 2020⁴⁸	NH	RTI, UTI, SSTI	Develop a decision support system to detect infections regardless of its type.
Rauh 2019⁵³	NH	RTI	1) Update a predictive model for 14-day mortality in antibiotic-treated NH residents with dementia and pneumonia; 2) risk stratification.
Steen 2006⁵⁴	NH	RTI	Develop and validate short- and mid-term mortality prediction models for antibiotic-treated NH residents with dementia and lower RTI.
Mehr 2001⁵⁰	NH	RTI	1) Identify patient characteristics predicting 30-day mortality in NH residents with lower RTI; 2) develop a simplified risk score to predict 30-day mortality; 3) risk stratification.
Naughton 2000⁵¹	NH	RTI	1) Develop a prediction model of 30-day mortality for NHAP; 2) apply the model to management issues related to NHAP; 3) risk stratification.

Open in a new tab

Notes: ML = machine learning; RTI = respiratory tract infection; UTI = urinary tract infections; SSTI = skin and soft tissue infections (wound infections); IV = intravenous; NHAP = nursing home-acquired pneumonia; HHC = home healthcare; NH = nursing home; ED = emergency department; RTLS = real-time location system.

Study quality evaluation

Figure 3 summarizes the quality appraisal results. All models exhibited low applicability concerns but a high overall risk of bias. Nine studies (69%) had a high risk in one domain,³⁴^,³⁵^,⁴⁷^,^51–53^,^55–57 three (23%) had a high or unclear risk in two domains,⁴⁹^,⁵⁰^,⁵⁴ and one had a high risk across all four domains.⁴⁸ The “Analysis” domain was the most problematic, with 92% of reviewed studies showing a high risk of bias.³⁴^,³⁵^,^47–52^,^54–57 Key issues in this domain included no external validation (92%),³⁴^,³⁵^,^47–53^,^55–57 no model calibration (69%),³⁴^,³⁵^,⁴⁸^,⁴⁹^,⁵¹^,⁵²^,^55–57 and inadequate handling of data complexities (54%), particularly multiple events per person and time-to-event considerations.³⁵^,^47–49^,⁵¹^,⁵⁶^,⁵⁷ Other deficiencies were a small outcome number (46%),³⁵^,⁴⁸^,⁴⁹^,^53–55 univariable predictor selection (38%),⁴⁷^,⁵⁰^,⁵⁴^,⁵⁶^,⁵⁷ not addressing missing data (31%),⁴⁹^,⁵¹^,⁵⁵^,⁵⁶ and a lack of details to prevent model overfitting and underfitting (31%).⁴⁷^,⁴⁹^,⁵²^,⁵⁶ In the “Predictor” domain (31%) (31%),⁴⁸^,⁵⁰^,⁵³^,⁵⁴ issues arose from predictor assessments not being independent of outcomes (23%)⁴⁸^,⁵⁰^,⁵³ and predictors not being uniformly assessed (15%).⁴⁸^,⁵⁴ “Outcome” issues (15%) included no standard outcome criteria (15%)⁴⁸^,⁴⁹ and outcome assessments not being independent of predictors (8%).⁴⁸ The issue of “Participants” (8%) was due to undefined inclusion/exclusion criteria.⁴⁸ See details in Supplementary Appendix 4.

Graph depicting the risk of bias assessment based on the Prediction Model Risk of Bias Assessment Tool criteria. Left shows the distribution of risk of bias results, while the right highlights common concerns related to risk of bias. — PROBAST risk of bias. A. Distribution of risk of bias results according to PROBAST criteria. B. Frequent risk of bias concerns. EPV=event per variable.

Data usage

All studies used structured EHR data, primarily from standardized assessments (85%)³⁴^,³⁵^,⁴⁷^,^50–57 and institution-specific EHR (92%).³⁴^,³⁵^,⁴⁷^,⁴⁸^,^50–57 Standardized assessments use uniform data elements for health evaluation, enabling easier comparisons across different settings.⁵⁸ We identified two main types: first are clinical assessment tools for specific conditions, primarily focusing on Activities of Daily Living (54%)³⁵^,⁴⁷^,⁵¹^,⁵²^,^55–57 and cognitive performance (46%).³⁵^,⁴⁷^,^52–54 Second are PAC routine assessments, which evaluate a broad range of conditions in specific PAC settings. For example, in the United States, the Minimum Data Set for NHs (23%)⁵⁰^,⁵²^,⁵⁴ and the Outcome and Assessment Information Set for HHC (15%)⁵⁶^,⁵⁷ are federally mandated at admission and discharge. Similarly, the Resident Assessment Instrument (interRAI) is used in Canadian long-term care facilities (15%).³⁴^,³⁵

Since 2020, 31% of studies incorporated additional data types, specifically bio-signal data from wearable devices,⁴⁸^,⁴⁹^,⁵⁵ and free-text data from clinical notes,⁵⁷ to develop multimodal models. Although the reviewed studies only used early fusion, multimodal models typically employ three fusion strategies-early, joint, or late fusion-to effectively integrate multiple data types.⁵⁹^,⁶⁰ Early (feature-level) fusion combines different modalities at the input level before feeding them into models.⁵⁹^,⁶⁰ Joint (intermediate-level) fusion first processes features from different modalities independently through separate networks, then merges them within intermediate neural network layers before final model training.⁵⁹^,⁶⁰ Late (decision-level) fusion integrates outputs from independent models to make a final prediction.⁵⁹^,⁶⁰

Data quality issues such as imbalance and missing data were commonly reported (Supplementary Appendix 6). Most studies (85%) noted a positive outcome rate below 30%,³⁴^,³⁵^,^49–57 while only 23% documented resampling procedures to achieve class balance.³⁴^,³⁵^,⁵⁵ This imbalance, coupled with a large number of predictors, often resulted in low events per variable ratio (46% studies lower than 10),³⁴^,⁴⁸^,⁴⁹^,^53–55 and risking model overfitting.¹⁹ The handling of missing data raised concern: 31% of studies did not report on their missing data practices;⁴⁷^,⁴⁹^,⁵²^,⁵⁶ only 8% used multiple imputation techniques,⁵⁴ while 38% resorted to single-value imputation,³⁴^,³⁵^,⁵⁰^,⁵³^,⁵⁴ and 23% to complete case analysis.⁵⁰^,⁵¹^,⁵⁷

Risk predictors

Figure 4 illustrates the distribution of risk predictors across socio-ecological levels, detailing specific predictor categories along with their respective data sources and types. All studies included individual-level predictors, primarily health conditions and symptoms (85%),³⁴^,³⁵^,⁴⁷^,^50–57 sociodemographic information (77%),³⁴^,³⁵^,⁴⁷^,⁵⁰^,^52–57 vital signs (62%),^48–54^,⁵⁶ and medical history and diagnosis (54%).³⁴^,⁴⁷^,⁵¹^,⁵²^,⁵⁴^,⁵⁶^,⁵⁷ A minority (15%) incorporated institutional-level predictors, such as facility capacity (bed numbers),³⁴^,³⁵ and community-level predictors, including community population density and the neighborhood-level deprivation index (Ontario Marginalization Index).³⁴^,³⁵ The least common were interpersonal-level predictors like living arrangements (8%),⁵⁶ and societal-level predictors such as public interest reflected through Google trends (8%).⁴⁸ Overall, the most commonly reported risk predictors were impaired cognitive function (62%),³⁵^,^51–57 declined Activities of Daily Living function (54%),³⁵^,^50–52^,^55–57 tachycardia (fast heart rate, 54%),^48–51^,⁵³^,⁵⁴^,⁵⁶ male sex (46%),³⁵^,⁵⁰^,^52–54^,⁵⁷ and skin ulcer (38%).³⁴^,³⁵^,⁵⁴^,⁵⁶^,⁵⁷ Supplementary Appendix 5 details risk predictors included in the finalized models, along with their data types and sources.

ML algorithms

ML algorithms were used in included studies for feature selection, model development and training (Table 1). For feature selection, 15% applied unsupervised techniques like deep temporal clustering,⁵⁵ and principal component analysis.⁴⁸ Regarding model development, most (69%) employed logistic regression, some (38%) used tree-based models such as random forest,⁴⁹^,⁵²^,⁵⁷ classification and regression trees,⁵⁵ and eXtreme Gradient Boosting (XgBoost).³⁴ The remaining studies utilized k-nearest neighbors⁴⁸^,⁴⁹ and neural network.⁵⁷ Notably, linear mixed-effects models⁵² and generalized estimating equations⁵⁰ were each used by 8% of the studies to manage the complexities of longitudinal data across PAC facilities and episodes for the same residents. Only one prognostic model adopted random survival forests for time-to-event analysis.³⁵ Furthermore, 23% of studies modified the learning process to prevent overfitting, such as shrinkage to penalize model complexity⁵⁴ and ensemble learning to combine predictions from models trained on different data subsets.³⁴^,³⁵ Supplementary Appendix 6 provided model development and evaluation details.

Validation techniques and evaluation metrics

External validation was lacking in 92% of studies,³⁴^,³⁵^,^47–53^,^55–57 with 38% relying solely on the holdout method for internal validation.³⁵^,⁴⁹^,⁵⁰^,⁵²^,⁵⁶ As a more robust alternative,¹⁹ 31% of studies used cross-validation,³⁴^,⁴⁸^,⁵⁷^,⁵⁷ and 38% applied bootstrapping to mitigate potential overfitting.⁴⁷^,⁴⁸^,^52–54 The evaluation measures used in the reviewed studies were highly diverse. Almost all (92%) studies reported discrimination metrics, notably the AUC/C-statistics (equivalent in binary classification tasks).³⁴^,³⁵^,⁴⁷^,⁴⁸^,^50–57 Other standard classification measures, such as accuracy, specificity, sensitivity, precision, F1 score, and net reclassification improvement, were reported in 46% of the studies.³⁴^,³⁵^,⁴⁸^,⁴⁹^,⁵⁵^,⁵⁷ Despite prevalent imbalanced outcomes, only one study reported the area under the precision-recall curve (AUPRC),⁵⁵ a metric specific to the imbalanced dataset. While model calibration is critical for comparing predictive probabilities with observed outcomes, only 38% of studies performed this assessment using the Hosmer-Lemeshow test and calibration plots.⁴⁷^,⁵⁰^,⁵³^,⁵⁴ Additionally, 15% of studies evaluated clinical utility using specialized measures like decision curve analysis, clinical impact curve, and face validity,⁴⁷^,⁵⁶ assessing the real-world usefulness and validity of the model predictions, especially in terms of the benefits and risks of clinical interventions based on model decisions. None of the reviewed studies applied fairness metrics, such as demographic parity⁶¹ and equal opportunity,⁶² to examine disparities in model performance stratified by demographic subgroups, socio-economic status, and contextual SDoH (e.g., geographic location, community socioeconomic status) across different socio-ecological levels.

Model performances

We focused on the best model performances within each study. For model discrimination, diagnostic models showed AUCs ranging from 0.69 to 0.958;³⁴^,⁴⁷^,⁴⁸^,^55–57 prognostic models had AUCs between 0.701 and 0.8.³⁵^,^50–54 For models on respiratory tract infections, AUCs ranged from 0.701 to 0.958, with diagnostic models performing higher (0.934 to 0.958)³⁴^,⁴⁷ than prognostic models (0.701 to 0.8).³⁵^,^50–54 The only models for urinary tract⁵⁵ and wound infections⁵⁷ were diagnostic, with AUCs of 0.69 and 0.818, respectively. The two remaining models, both diagnostic for various infections together, showed AUCs of 0.7162 and 0.798.⁴⁸^,⁵⁶

Table 2 presents the highest reported performance metrics across studies (e.g., AUC, sensitivity, precision). These metrics may come from different models within each study. The best-performing ML algorithm varied across studies and depended on the evaluation metrics used. No significant differences in discrimination were observed based on the modeling approach, whether in terms of algorithms (regression vs non-regression) or data modality (multimodal vs unimodal). See Supplementary Appendix 6—Table S2 for other model evaluation details.

Table 2.

Model performance by ML algorithms and data modality.

Study	ML algorithms	Multimodal	Best performance
Lee 2022³⁴	XgBoost		C-statistics: 0.934 (95% CI: 0.915-0.951) Sensitivity: 0.887 Specificity: 0.869
Ramazi 2022⁵⁵	Deep temporal clustering Classification and regression tree	●	AUC: 0.69 Sensitivity: 0.91 ± 0.09 Specificity: 0.71 ± 0.04 Precision: 0.76 ± 0.04 AUPRC: 0.70
Tan 2022⁴⁷	Logistic regression		C-statistics: 0.958 (95% CI: 0.943-0.972)
Garcés 2021⁴⁹	K-nearest neighbors Random forest	●	Accuracy: 1.00
Lee 2021³⁵	Random forest survival model		AUC: 0.701 (95% CI: 0.666, 0.736)
Song 2021⁵⁷	Logistic regression Random forest Neural network	●	AUC: 0.818 Sensitivity: 0.876 Accuracy: 0.750 Specificity: 0.749 F1-score: 0.88
Panagiotou 2021⁵²	Logistic regression		AUC: 0.74 (95% CI, 0.73-0.77)
Shang 2020⁵⁶	Logistic regression		C-statistics: 0.7162
Baldominos 2020⁴⁸	K-nearest neighbors Principal component analysis	●	AUC: 0.798 Sensitivity: 0.723 Precision: 0.777 F1 score: 0.734
Rauh 2019⁵³	Logistic regression		AUC: 0.80 (IQR: 0.80–0.81)
Steen 2006⁵⁵	Logistic regression		C-statistics: 0.74
Mehr 2001⁵⁰	Logistic regression		C-statistics: 0.76
Naughton 2000⁵¹	Logistic regression		AUC: 0.74

Open in a new tab

Notes: AUC= area under the curve (equivalent to C-statistics= concordance index in binary classification); AUPRC= The area under Precision-Recall curve; CI= confidence interval; IQR: inter-quartile range; ● = Yes.

Model interpretability

Model interpretability is crucial for providing insights into influential factors and supporting clinical decision-making. All reviewed studies used interpretability methods. Built-in interpretability methods were predominantly used (77%),^47–54^,⁵⁶^,⁵⁷ deriving from inherently interpretable models such as logistic regression and tree-based models. The remaining 23% of studies utilized post-hoc interpretability methods: one study utilized permutation methods to visualize the 50 most important predictors;³⁵ SHapley Additive exPlanations (SHAP), a more advanced model explainability method based on cooperative game theory, was adopted by two studies.³⁴^,⁵⁵

Model translation and clinical applications

Reviewed studies highlight three main ML applications in infection-related risk prediction in PAC: (1) developing individualized predictive models to support tailoring clinical interventions (100%),³⁴^,³⁵^,^47–57 (2) conducting risk stratification to help allocate healthcare resources effectively (54%),³⁴^,³⁵^,^50–53^,⁵⁶ and (3) identifying new risk predictors to reveal new solutions for infection prevention and control (46%).³⁴^,³⁵^,⁵⁰^,⁵²^,⁵⁵^,⁵⁷

To translate the predictive risk models into practical clinical tools, most studies (61%) converted probabilistic outputs into estimated classes, providing binary or multi-class predictions.³⁴^,³⁵^,⁴⁸^,⁴⁹^,⁵²^,^55–57 A few studies (38%) translated the underlying multivariable logistic regression models into additive scoring systems by assigning integer values to finalized risk predictors.⁴⁷^,⁵⁰^,⁵¹^,⁵³^,⁵⁴ Although the simplified scoring systems were straightforward, bias issues were noted during the translation process, as highlighted by PROBAST.¹⁹ Specifically, there was a mismatch where some predictors were removed during the translation and/or the assigned points did not correspond accurately to the coefficients from the finalized multivariable models.⁵¹^,⁵⁴ This mismatch can lead to misinterpretations of the predictor contributions to the risk and consequently distort the estimated individual risk based solely on assigned points.¹⁹ Some studies (54%) stratified risk into 4–6 quartiles to illustrate how infection risk increases across different levels.³⁴^,³⁵^,^50–53^,⁵⁶ No research team conducted subsequent external validation or deployed the translated model products (e.g., simplified risk scoring systems) into real-world clinical practice beyond the original development studies.

Discussion

This is the first review to critically analyze published ML-based diagnostic and prognostic models for PAC settings, emphasizing their effectiveness in identifying infection-related risks. Primarily, structured EHR data were used, occasionally supplemented with bio-signals from wearable devices or free text from clinical notes. Conventional statistical and ML models were predominantly used, whereas deep learning model utilization was limited. The top methodological challenges include a lack of external validation, no calibration, and inadequate consideration of data complexities. Figure 5 summarizes the barriers identified in the included studies that need to be addressed regarding underlying data usage, risk predictor measuring and integrating, model derivation and evaluation, and clinical translation.

A graph showing barriers in machine learning-based infection diagnostic and prognostic models for post-acute care, including challenges in data usage, risk predictor integration, model development, and clinical translation. — Barriers in ML-based infection diagnostic and prognostic models for PAC.

Overall, this review reveals a focus on NHs and less exploration of other PAC settings. There is an imbalance in ML applications for different infection types, particularly the absence of sepsis models, despite its major role in hospital transfers.⁶³ Reviewed studies focused more on type-specific risk prediction models and lacked generic infection models. This finding consistent with another literature review of older adults’ infection prediction.²³ However, our results noted small positive outcome classes in specific infection types and shared atypical symptoms and underlying comorbidities across different infection types. Given these findings, more research on generic infection models should be considered, and compare with type-specific infection models to examine their potential. Additionally, feedback from clinicians is critical for determining which models can better inform clinical decision-making.

Results show a growing diversity in data sources and types for infection-related risk prediction in PAC. Since 2020, half of the studies have expanded beyond structured EHR data,⁴⁸^,⁴⁹^,⁵⁷ with the support of advanced feature extraction and fusion techniques to utilize multi-sourced unstructured data.⁵⁵^,⁵⁷ Similar trends are also noted in other infection-related risk prediction reviews.²²^,²⁴ With the support of fusion strategies, this approach can help identify the interactions between various data elements and enhance patient profiling.²¹^,²⁹^,⁶⁴ However, further demonstrating the superiority of multimodal model performance in this review is challenging. This may be due to the limited number of multimodal models available for pooled performance analysis and the high variability in study designs affecting outcomes such as sample size, predictor combinations, and events per variable.¹⁹ Most studies lacked direct model comparisons. With ongoing feature extraction and risk modeling advancements, future research is encouraged to identify the optimal data modalities and ML model usage in PAC settings, including effective fusion strategies.

Despite its potential, achieving data multimodality in PAC research faces significant challenges, particularly in data collection. Compared to acute-care settings, which are well-suited to using pathology, imaging, and omics data for predictive modeling,⁶⁵ PAC settings have limited access to these resources due to its healthcare nature, focusing instead on regaining function to achieve daily self-sufficiency.⁶⁶ This review highlights the potential of wearable devices to capture functional impairments and clinical indicators crucial for infection prediction in PAC.⁵⁵ It is recommended that future research explore additional data collection opportunities through biosensors and ambient intelligence technologies, such as acoustic sensors for respiratory infection detection⁶⁷ and thermal sensors for wound infection detection,⁶⁸ with sufficient consideration of privacy and ethical concerns.

PAC settings, being less intensive and more connected to the external environment, necessitate a broader socio-ecological lens. While a large body of ML research targets individual-level predictors, this review underlines the importance of incorporating contextual-level predictors guided by the socio-ecological conceptual framework. These include interpersonal predictors like living arrangements,⁵⁶ institutional predictors like bed numbers, community characteristics like neighborhood deprivation indices,³⁴^,³⁵ and societal influences like public interest in infection topics.⁴⁸ Further research should explore how additional contextual-level predictors—such as living environments, social norms, and health policies—to better understand their impact on infection-related health outcomes, thereby providing a more comprehensive understanding.

This review highlights significant concerns about algorithmic fairness, particularly the inadequate reporting of sociodemographic and SDoH data in most studies. This omission hinders the ability to assess variability and ensure representations across diverse subgroups. Moreover, none of the reviewed studies evaluated model performance across different demographic groups or employed fairness metrics to identify disparities. To ensure a fair risk prediction model, further research should adopt a comprehensive strategy to facilitate measuring, reporting, and integrating SDoH. First, it is vital to thoroughly report demographics and SDoH data to achieve diverse dataset representation.⁶⁹ Second, as defined by the WHO,⁷⁰ SDoH encompasses a broad range of factors beyond individual demographics, including the conditions in which people are born, grow, live, work, and age. Therefore, there is a need for improved measurement and integration of these contextual-level SDoH within the socio-ecological conceptual framework.⁴⁰^,⁷¹ For example, advances in natural language processing and computer vision offer promising solutions for the scalable extraction of SDoH features from unstructured data, including those related to physical and social living environments.⁷²^,⁷³ Third, incorporating fairness metrics in model evaluation is crucial to ensure accurate predictions for diverse subgroups.

While advanced ML models offer greater potential for handling large multimodal data, and improving performance over traditional regression models regarding infection-related risk modeling in PAC, more emphasis should be placed on their clinical utility pitfalls. As trends shift towards multimodal ML, the capability of a model often increases, yet at the cost of time-consuming data concatenation and development processes.⁷⁴ These complexities hinder the understanding of predictive rationales, reducing interpretability.

In reviewed studies, additive risk-scoring systems derived from conventional models with around 10 predictors⁴⁹^,⁵⁰^,⁵¹^,⁵³^,⁵⁴ have shown more potential as clinical tools than those with many predictors. These complex models, often featuring advanced algorithms, may inhibit clinical usefulness by increasing the cognitive load for clinicians.⁷⁵ For example, assessing a patient might require interpreting data from over 100 predictors to make a clinical decision. However, advances in interpretability methods, such as SHAP, enhance the understanding of predictor contributions by providing quantitative and visually intuitive insights that support clinicians in practical settings.³⁴^,⁷⁶ This progress encourages the continued exploration of sophisticated ML algorithms in PAC settings without overburdening clinicians. Nevertheless, further research is needed into advanced explainable AI methods, particularly for unstructured data,⁷⁷ to provide clinically meaningful and actionable insights.

A critical gap in assessing the clinical utility of ML models was also revealed. Despite some studies evaluating clinical utility through expert agreement and impact assessment,⁴⁷^,⁵⁶ it is argued that relying solely on internal and external validation is insufficient.⁷⁸ The absence of robust evidence regarding the real-world benefits of ML algorithms necessitates caution when interpreting them as effective clinical interventions. Future research should also consider model deployment feasibility, such as determining the optimal time window and feature size for data censoring and extraction. Collaboration among developers, clinicians, and managers is essential for achieving this goal.

This review highlights several methodological flaws hindering the progress and implementation of these models, notably the lack of external validation. This ongoing issue questions their generalizability and contributes to a replication crisis in model applicability.⁷⁹ Furthermore, underreporting model calibration, crucial for performance assessment, means uncalibrated models may not perform well in real-world scenarios.⁸⁰ Such a lack of detailed performance reporting hinders a proper model assessment and comparison. Given this gap, we advocate for building standardized dataset constructs and using common data models to enhance external validity.⁸¹ A common data model standardizes biomedical data representation by aligning entities, attributes, and relationships across multiple sources.⁸² Notable examples include the Observational Medical Outcomes Partnership, the National Patient-Centered Clinical Research Network, and Sentinel.⁸³ This standardization facilitates external validation by integrating consistent data from multiple sources into a harmonized dataset, ensuring that ML model inferences are meaningfully comparable across different studies.⁸²^,⁸³ Comprehensive performance evaluation is also recommended. Apart from standard discrimination measures, calibration measures such as calibration plots and the Hosmer-Lemeshow test should be incorporated.

This review also emphasizes the need for more appropriate modeling approaches to address data complexity. Traditional data independence assumptions are unsuitable for PAC contexts with interrelated episodes for the same residents. Promising alternatives for further studies may lie in causal inference and graph neural networks (under the deep learning umbrella), which account for data point interconnections.⁸⁴^,⁸⁵ Beyond traditional resampling methods to address data imbalance, advancements in large language models may effectively synthesize multimodal data profiles to depict underrepresented groups.⁸⁶ Furthermore, advanced models like Derep Surv,⁸⁷ Cox-nnet,⁸⁸ and DeepHit,⁸⁹ are recommended for managing extensive, complex time-to-event data.

Limitations

Despite a comprehensive search across six databases and hand searches, this review may have overlooked current prediction models due to excluding most grey literature. Additional limitations include (1) potential overestimation of model effectiveness due to reporting only the best algorithm performances; (2) exclusion of studies published in languages other than English; (3) significant heterogeneity in data sources, study populations, and predictor combinations, complicating the interpretation and comparison of model performance across different tasks, and potentially impacting the representativeness of the results, especially when few studies are available; and (4) exclusion of studies that only evaluated existing models without updating predictors or algorithms, potentially missing relevant findings and limiting the generalizability of our results across different sites.

Conclusions

This systematic review analyzed ML-based infection diagnostic and prognostic models in PAC settings. The literature indicates a shift toward multimodal risk models and advanced modeling approaches. However, we found limited evidence of their superiority over traditional regression models using structured EHR data due to high heterogeneity and low generalizability on validation sets. Challenges identified include PAC data complexity, compromised clinical utility of models, and the absence of external validity. Despite these limitations, we recommend continuing ML applications in infection-related risk predictions at PAC, integrating multi-level risk predictors from diverse data sources, and emphasizing efficient multimodal feature analysis and model interpretability.

Supplementary Material

ocae278_Supplementary_Data

ocae278_supplementary_data.docx^{(65.3KB, docx)}

Acknowledgments

We thank John Usseglio at Columbia University for the assistance with the literature search. We would also like to thank Drs. Gregory L. Alexander, Corina Lelutiu-Weinberger, Patricia W. Stone and Arlene Smaldone for their valuable suggestions to our manuscript.

Contributor Information

Zidu Xu, School of Nursing, Columbia University, New York, NY 10032, United States.

Danielle Scharp, School of Nursing, Columbia University, New York, NY 10032, United States.

Mollie Hobensack, Icahn School of Medicine at Mount Sinai, New York, NY 10029, United States.

Jiancheng Ye, Weill Cornell Medicine, Cornell University, New York, NY 10065, United States.

Jungang Zou, Department of Biostatistics, Mailman School of Public Health, Columbia University, New York, NY 10032, United States.

Sirui Ding, Bakar Computational Health Sciences Institute, University of California, San Francisco, CA 94158, United States.

Jingjing Shang, School of Nursing, Columbia University, New York, NY 10032, United States.

Maxim Topaz, School of Nursing, Columbia University, New York, NY 10032, United States; Center for Home Care Policy & Research, VNS Health, New York, NY 10001, United States; Data Science Institute, Columbia University, New York, NY 10027, United States.

Supplementary material

Supplementary material is available at Journal of the American Medical Informatics Association online.

Author contributions

ZX conceptualized the study design and research questions, with input from MT and JS. ZX searched the electronic databases and conducted backward and forward reference list checking. ZX, DS, MH, JY, and JZ performed screening, study selection, and data extraction. ZX performed data synthesis and contributed to the original draft, with input from SD. MT, JS and JY performed review and editing. MT and JS supervised the study. All authors approved the final draft for submission.

Funding

Agency for Healthcare Research and Quality (AHRQ) grant numbers R01 HS027742 (to MT), R01 HS028637 (to JS); National Institute of Nursing Research (NINR) grant numbers 2R01NR016865 (to JS), T32NR007969 (to DS); 2024 Marilyn D. Harris Research Grant (to ZX); 2023 Sigma Small Grants (to ZX); 2024 Home Care Dissertation Research Grant (to ZX).

Conflicts of interest

MT and JS are the authors of 2 articles in the systematic review. In order to address the conflict of interest and reduce bias, DS and JZ assisted ZX with the review of those studies independently.

Data availability

All data generated during this study are provided as supplementary materials.

References

1. The Medicare Payment Advisory Commission. Chapter 8. In July 2023 Data Book: Health Care Spending and the Medicare Program. The Medicare Payment Advisory Commission, 2023.
2. Wang YC, Chou MY, Liang CK, Peng LN, Chen LK, Loh CH. Post-acute care as a key component in a healthcare system for older adults. Ann Geriatr Med Res. 2019;23:54-62. 10.4235/agmr.19.0009 [DOI] [PMC free article] [PubMed] [Google Scholar]
3. Centers for Medicare & Medicaid Services. Medicare and Medicaid Programs: Revisions to Requirements for Discharge Planning for Hospitals, Critical Access Hospitals, and Home Health Agencies. Federal Register. Accessed March 9, 2024. https://www.federalregister.gov/documents/2019/09/30/2019-20732/medicare-and-medicaid-programs-revisions-to-requirements-for-discharge-planning-for-hospitals
4. National Center for Health Statistics. Biennial Overview of Post-Acute and Long-Term Care in the United States. Centers for Disease Control and Prevention. Accessed January 8, 2024. https://data.cdc.gov/d/wibz-pb5q
5. Valiathan R, Ashman M, Asthana D. Effects of ageing on the immune system: infants to elderly. Scand J Immunol. 2016;83:255-266. 10.1111/sji.12413. [DOI] [PubMed] [Google Scholar]
6. Gavazzi G, Krause K-H. Ageing and infection. Lancet Infect Dis. 2002;2:659-666. 10.1016/S1473-3099(02)00437-1. [DOI] [PubMed] [Google Scholar]
7. El Chakhtoura NG, Bonomo RA, Jump RLP. Influence of Aging and Environment on Presentation of Infection in Older Adults. Infect Dis Clin North Am. 2017;31:593-608. 10.1016/j.idc.2017.07.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
8. Smith LA, Oakden-Rayner L, Bird A, et al. Machine learning and deep learning predictive models for long-term prognosis in patients with chronic obstructive pulmonary disease: a systematic review and meta-analysis. Lancet Digit Health. 2023;5:e872-e81. 10.1016/s2589-7500(23)00177-2 [DOI] [PubMed] [Google Scholar]
9. Kline KA, Bowdish DM. Infection in an aging population. Curr Opin Microbiol. 2016;29:63-67. 10.1016/j.mib.2015.11.003 [DOI] [PubMed] [Google Scholar]
10. Mitchell SL, Teno JM, Kiely DK, et al. The clinical course of advanced dementia. N Engl J Med. 2009;361:1529-1538. 10.1056/NEJMoa0902234 [DOI] [PMC free article] [PubMed] [Google Scholar]
11. Givens JL, Selby K, Goldfeld KS, Mitchell SL. Hospital transfers of nursing home residents with advanced dementia. J Am Geriatr Soc. 2012;60:905-909. 10.1111/j.1532-5415.2012.03919.x [DOI] [PMC free article] [PubMed] [Google Scholar]
12. Ouslander JG, Bonner A, Herndon L, Shutes J. The Interventions to Reduce Acute Care Transfers (INTERACT) quality improvement program: an overview for medical directors and primary care clinicians in long term care. J Am Med Dir Assoc. 2014;15:162-170. 10.1016/j.jamda.2013.12.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
13. Tingström P, Milberg A, Rodhe N, Ernerud J, Grodzinsky E, Sund-Levander M. Nursing assistants: “he seems to be ill” - a reason for nurses to take action: validation of the Early Detection Scale of Infection (EDIS). BMC Geriatr. 2015;15:122. 10.1186/s12877-015-0114-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
14. Vogenberg FR. Predictive and prognostic models: implications for healthcare decision-making in a modern recession. Am Health Drug Benefits. 2009;2:218-222. [PMC free article] [PubMed] [Google Scholar]
15. Kawamoto K, McDonald CJ. Designing, conducting, and reporting clinical decision support studies: recommendations and call to action. Ann Intern Med. 2020;172:S101-S109. 10.7326/m19-0875 [DOI] [PubMed] [Google Scholar]
16. Andaur Navarro CL, Damen JAA, van Smeden M, et al. Systematic review identifies the design and methodological conduct of studies on machine learning-based prediction models. J Clin Epidemiol. 2023;154:8-22. 10.1016/j.jclinepi.2022.11.015. [DOI] [PubMed] [Google Scholar]
17. Theodosiou AA, Read RC. Artificial intelligence, machine learning and deep learning: Potential resources for the infection clinician. J Infect. 2023;87:287-294. [DOI] [PubMed] [Google Scholar]
18. Jordan MI, Mitchell TM. Machine learning: Trends, perspectives, and prospects. Science. 2015;349:255-260. 10.1126/science.aaa8415 [DOI] [PubMed] [Google Scholar]
19. Moons KGM, Wolff RF, Riley RD, et al. PROBAST: a tool to assess risk of bias and applicability of prediction model studies: explanation and elaboration. Ann Intern Med. 2019;170:W1-W33. 10.7326/m18-1377 [DOI] [PubMed] [Google Scholar]
20. Moons KG, de Groot JA, Bouwmeester W, et al. Critical appraisal and data extraction for systematic reviews of prediction modelling studies: the CHARMS checklist. PLoS Med. 2014;11:e1001744. 10.1371/journal.pmed.1001744 [DOI] [PMC free article] [PubMed] [Google Scholar]
21. Xu M, Ouyang L, Han L, et al. Accurately differentiating between patients with COVID-19, patients with other viral infections, and healthy individuals: multimodal late fusion learning approach. J Med Internet Res. 2021;23:e25535. 10.2196/25535 [DOI] [PMC free article] [PubMed] [Google Scholar]
22. Ren H, Ling Y, Cao R, Wang Z, Li Y, Huang T. Early warning of emerging infectious diseases based on multimodal data. Biosaf Health. 2023;5:229-236. 10.1016/j.bsheal.2023.05.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
23. Masot O, Cox A, Mold F, et al. Decision support-tools for early detection of infection in older people (aged> 65 years): a scoping review. BMC Geriatr. 2022;22:552. 10.1186/s12877-022-03218-w [DOI] [PMC free article] [PubMed] [Google Scholar]
24. Yan MY, Gustad LT, Nytrø Ø. Sepsis prediction, early detection, and identification using clinical text for machine learning: a systematic review. J Am Med Inform Assoc. 2022;29:559-575. [DOI] [PMC free article] [PubMed] [Google Scholar]
25. Goh KH, Wang L, Yeow AYK, et al. Artificial intelligence in sepsis early prediction and diagnosis using unstructured data in healthcare. Nat Commun. 2021;12:711. 10.1038/s41467-021-20910-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
26. Enshaeifar S, Zoha A, Skillman S, et al. Machine learning methods for detecting urinary tract infection and analysing daily living activities in people with dementia. PLoS One. 2019;14:e0209909. 10.1371/journal.pone.0209909 [DOI] [PMC free article] [PubMed] [Google Scholar]
27. Stokes K, Castaldo R, Federici C, et al. The use of artificial intelligence systems in diagnosis of pneumonia via signs and symptoms: A systematic review. Biomedical Signal Processing and Control. 2022;72:103325. 10.1016/j.bspc.2021.103325. [DOI] [Google Scholar]
28. Zhu C, Xu Z, Gu Y, et al. Prediction of post-stroke urinary tract infection risk in immobile patients using machine learning: an observational cohort study. J Hosp Infect. 2022;122:96-107. 10.1016/j.jhin.2022.01.002. [DOI] [PubMed] [Google Scholar]
29. Fang C, Yan W, Chen Y, et al. Long-term SARS-CoV-2 neutralizing antibody level prediction using multimodal deep learning: A prospective cohort study on longitudinal data in Wuhan, China. J Med Virol. 2023;95:e29036. 10.1002/jmv.29036. [DOI] [PubMed] [Google Scholar]
30. Adams V, Song J, Shang J, et al. Infection prevention and control practices in the home environment: Examining enablers and barriers to adherence among home health care nurses. Am J Infect Control. 2021;49:721-726. 10.1016/j.ajic.2020.10.021. [DOI] [PMC free article] [PubMed] [Google Scholar]
31. Strupp J, Valerius L, Herrler A, et al. I see the high level of interpersonal support during the pandemic as proof that the good in people prevails': the COVID-19 pandemic as a catalyst for building a caring community? J Public Health (Oxf). 2024;46:175-184. 10.1093/pubmed/fdad283 [DOI] [PubMed] [Google Scholar]
32. Katz MJ, Roghmann MC. Healthcare-associated infections in the elderly: what's new. Curr Opin Infect Dis. 2016;29:388-393. 10.1097/qco.0000000000000283 [DOI] [PMC free article] [PubMed] [Google Scholar]
33. Siegel EO, Backman A, Cai Y, et al. Understanding contextual differences in residential LTC provision for cross-national research: identifying internationally relevant CDEs. Gerontol Geriatr Med. 2019;5:2333721419840591. 10.1177/2333721419840591 [DOI] [PMC free article] [PubMed] [Google Scholar]
34. Lee DS, Wang CX, McAlister FA, et al. Factors associated with SARS-CoV-2 test positivity in long-term care homes: a population-based cohort analysis using machine learning. The Lancet Regional Health—Americas. 2022;6:100146. 10.1016/j.lana.2021.100146 [DOI] [PMC free article] [PubMed] [Google Scholar]
35. Lee DS, Ma S, Chu A, CORONA Collaboration, et al. Predictors of mortality among long-term care residents with SARS-CoV-2 infection. Journal of the American Geriatrics Society. 2021;69:3377-3388. 10.1111/jgs.17425. [DOI] [PubMed] [Google Scholar]
36. Stone PW, Herzig CTA, Agarwal M, Pogorzelska-Maziarz M, Dick AW. Nursing home infection control program characteristics, CMS citations, and implementation of antibiotic stewardship policies: a national study. Inquiry. 2018;55:46958018778636. 10.1177/0046958018778636 [DOI] [PMC free article] [PubMed] [Google Scholar]
37. Houben F, Heijer CDD, Dukers-Muijrers NH, Smeets-Peels C, Hoebe CJ. Psychosocial determinants associated with healthcare workers' self-reported compliance with infection prevention and control during the COVID-19 pandemic: a cross-sectional study in Dutch residential care facilities for people with intellectual and developmental disabilities. BMC Public Health. 2023;23:2052. 10.1186/s12889-023-16912-0 [DOI] [PMC free article] [PubMed] [Google Scholar]
38. Kilanowski JF. Breadth of the socio-ecological model. J Agromedicine. 2017;22:295-297. 10.1080/1059924X.2017.1358971 [DOI] [PubMed] [Google Scholar]
39. Bronfenbrenner U. Ecology of the family as a context for human development: Research perspectives. Developmental Psychology. 1986;22:723-742. 10.1037/0012-1649.22.6.723 [DOI] [Google Scholar]
40. Mhasawade V, Zhao Y, Chunara R. Machine learning and algorithmic fairness in public and population health. Nat Mach Intell. 2021;3:659-666. [Google Scholar]
41. Matthew JP, Joanne EM, Patrick MB, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. 10.1136/bmj.n71 [DOI] [PMC free article] [PubMed] [Google Scholar]
42. Badillo S, Banfai B, Birzele F, et al. An introduction to machine learning. Clin Pharmacol Ther. 2020;107:871-885. 10.1002/cpt.1796. [DOI] [PMC free article] [PubMed] [Google Scholar]
43. Goldstein BA, Navar AM, Carter RE. Moving beyond regression techniques in cardiovascular risk prediction: applying machine learning to address analytic challenges. Eur Heart J. 2017;38:1805-1814. 10.1093/eurheartj/ehw302 [DOI] [PMC free article] [PubMed] [Google Scholar]
44. Bramer WM, Giustini D, de Jonge GB, Holland L, Bekhuis T. De-duplication of database search results for systematic reviews in EndNote. J Med Libr Assoc. 2016;104:240-243. 10.3163/1536-5050.104.3.014 [DOI] [PMC free article] [PubMed] [Google Scholar]
45. Wolff RF, Moons KGM, Riley RD, PROBAST Group†, et al. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies. Annals of Internal Medicine. 2019;170:51-58. 10.7326/M18-1376 [DOI] [PubMed] [Google Scholar]
46. Pham MT, Rajić A, Greig JD, Sargeant JM, Papadopoulos A, McEwen SA. A scoping review of scoping reviews: advancing the approach and enhancing the consistency. Res Synth Methods. 2014;5:371-385. 10.1002/jrsm.1123 [DOI] [PMC free article] [PubMed] [Google Scholar]
47. Tan PY, Huo M, Zhou XH, Zhao BL. Development and validation of a nomogram for predicting the risk of nursing home-acquired pneumonia. Eur Rev Med Pharmacol Sci. 2022;26:8276-8288. 10.26355/eurrev_202211_30360 [DOI] [PubMed] [Google Scholar]
48. Baldominos A, Ogul H, Colomo-Palacios R, Sanz-Moreno J, Gómez-Pulido JM. Infection prediction using physiological and social data in social environments. Information Processing & Management. 2020;57:102213. 10.1016/j.ipm.2020.102213. [DOI] [Google Scholar]
49. Garcés-Jiménez A, Calderón-Gómez H, Gómez-Pulido JM, et al. Medical prognosis of infectious diseases in nursing homes by applying machine learning on clinical data collected in cloud microservices. Int J Environ Res Public Health. 2021;18 [DOI] [PMC free article] [PubMed] [Google Scholar]
50. Mehr DR, Binder EF, Kruse RL, et al. Predicting mortality in nursing home residents with lower respiratory tract infection: the Missouri LRI study. JAMA. 2001;286:2427-2436. 10.1001/jama.286.19.2427 [DOI] [PubMed] [Google Scholar]
51. Naughton BJ, Mylotte JM, Tayara A. Outcome of nursing home-acquired pneumonia: derivation and application of a practical model to predict 30 day mortality. J Am Geriatr Soc. 2000;48:1292-1299. 10.1111/j.1532-5415.2000.tb02604.x. [DOI] [PubMed] [Google Scholar]
52. Panagiotou OA, Kosar CM, White EM, et al. Risk factors associated with all-cause 30-day mortality in nursing home residents with COVID-19. JAMA Intern Med. 2021;181:439-448. 10.1001/jamainternmed.2020.7968 [DOI] [PMC free article] [PubMed] [Google Scholar]
53. Rauh SP, Heymans MW, van der Maaden T, et al. Predicting mortality in nursing home residents with dementia and pneumonia treated with antibiotics: validation of a prediction model in a more recent population. J Gerontol A Biol Sci Med Sci. 2019;74:1922-1928. 10.1093/gerona/gly260 [DOI] [PubMed] [Google Scholar]
54. van der Steen JT, Mehr DR, Kruse RL, et al. Predictors of mortality for lower respiratory infections in nursing home residents with dementia were validated transnationally. J Clin Epidemiol. 2006;59:970-979. 10.1016/j.jclinepi.2005.12.005. [DOI] [PubMed] [Google Scholar]
55. Ramazi R, Bowen ME, Flynn AJ, Beheshti R. Developing acute event risk profiles for older adults with dementia in long-term care using motor behavior clusters derived from deep learning. J Am Med Dir Assoc. 2022;23:1977-1983.e1. 10.1016/j.jamda.2022.04.009. [DOI] [PubMed] [Google Scholar]
56. Shang J, Russell D, Dowding D, et al. A Predictive risk model for infection-related hospitalization among home healthcare patients. J Healthc Qual. 2020;42:136-147. [DOI] [PMC free article] [PubMed] [Google Scholar]
57. Song J, Woo K, Shang J, Ojo M, Topaz M. Predictive risk models for wound infection-related hospitalization or ED visits in home health care using machine-learning algorithms. Adv Skin Wound Care. 2021;34:1-12. 10.1097/01.Asw.0000755928.30524.22 [DOI] [PubMed] [Google Scholar]
58. Maritz R, Ehrmann C, Prodinger B, Tennant A, Stucki G. The influence and added value of a Standardized Assessment and Reporting System for functioning outcomes upon national rehabilitation quality reports. Int J Qual Health Care. 2020;32:379-387. 10.1093/intqhc/mzaa058 [DOI] [PMC free article] [PubMed] [Google Scholar]
59. Mohsen F, Ali H, El Hajj N, Shah Z. Artificial intelligence-based methods for fusion of electronic health records and imaging data. Sci Rep. 2022;12:17981. 10.1038/s41598-022-22514-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
60. Huang S-C, Pareek A, Seyyedi S, Banerjee I, Lungren MP. Fusion of medical imaging and electronic health records using deep learning: a systematic review and implementation guidelines. NPJ Digit Med. 2020;3:136. 10.1038/s41746-020-00341-z [DOI] [PMC free article] [PubMed] [Google Scholar]
61. Calders T, Kamiran F, Pechenizkiy M. Building classifiers with independency constraints. In: 2009 IEEE International Conference on Data Mining Workshops, Miami, FL, USA. IEEE Computer Society; 2009:13-18.. 10.1109/ICDMW.2009.83 [DOI]
62. Hardt M, Price E, Srebro N. Equality of opportunity in supervised learning. Proceedings of the 30th International Conference on Neural Information Processing Systems. Curran Associates Inc., 2016:3323-31.
63. Harrison JM, Dick AW, Stone PW, et al. Infection trends in home health care, 2013–2018. Infect Control Hosp Epidemiol. 2021;42:1388-1390. 10.1017/ice.2021.248 [DOI] [PMC free article] [PubMed] [Google Scholar]
64. Naegelin M, Weibel RP, Kerr JI, et al. An interpretable machine learning approach to multimodal stress detection in a simulated office environment. J Biomed Inform. 2023;139:104299. 10.1016/j.jbi.2023.104299. [DOI] [PubMed] [Google Scholar]
65. Coudray N, Ocampo PS, Sakellaropoulos T, et al. Classification and mutation prediction from non–small cell lung cancer histopathology images using deep learning. Nat Med. 2018;24:1559-1567. 10.1038/s41591-018-0177-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
66. Association AH. Fact Sheet: Post-acute Care. Secondary Fact Sheet: Post-acute Care 2019. https://www.aha.org/system/files/media/file/2019/07/fact-sheet-post-acute-care-0719.pdf.
67. Pinkas G, Karny Y, Malachi A, Barkai G, Bachar G, Aharonson V. SARS-CoV-2 detection from voice. IEEE Open J Eng Med Biol. 2020;1:268-274. 10.1109/ojemb.2020.3026468 [DOI] [PMC free article] [PubMed] [Google Scholar]
68. Xiong Z, Achavananthadith S, Lian S, et al. A wireless and battery-free wound infection sensor based on DNA hydrogel. Sci Adv. 2021;7:eabj1617. 10.1126/sciadv.abj1617 [DOI] [PMC free article] [PubMed] [Google Scholar]
69. Centers for Medicare & Medicaid Services. The CMS Framework for Health Equity 2022-2032. 2022. https://www.cms.gov/files/document/cms-framework-health-equity-2022.pdf
70. WHO Regional Office for South-East Asia. Social Determinants of Health. 2008
71. Yearby R. Structural racism and health disparities: reconfiguring the social determinants of health framework to include the root cause. J Law Med Ethics. 2020;48:518-526. [DOI] [PubMed] [Google Scholar]
72. Patra BG, Sharma MM, Vekaria V, et al. Extracting social determinants of health from electronic health records using natural language processing: a systematic review. J Am Med Inform Assoc. 2021;28:2716-2727. 10.1093/jamia/ocab170 [DOI] [PMC free article] [PubMed] [Google Scholar]
73. Rehman N, Saif U, Chunara R. Deep landscape features for improving vector-borne disease prediction. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops. IEEE; 2019:44-51.
74. Kline A, Wang H, Li Y, et al. Multimodal machine learning in precision health: A scoping review. NPJ Digit Med. 2022;5:171. 10.1038/s41746-022-00712-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
75. Van Calster B, Wynants L, Timmerman D, Steyerberg EW, Collins GS. Predictive analytics in health care: how can we know it works? J Am Med Inform Assoc. 2019;26:1651-1654. 10.1093/jamia/ocz130 [DOI] [PMC free article] [PubMed] [Google Scholar]
76. Lundberg SM, Lee S-I. A unified approach to interpreting model predictions. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. Curran Associates Inc.; 2017:4768-4777. [Google Scholar]
77. Payrovnaziri SN, Chen Z, Rengifo-Moreno P, et al. Explainable artificial intelligence models using real-world electronic health record data: a systematic scoping review. J Am Med Inform Assoc. 2020;27:1173-1185. 10.1093/jamia/ocaa053 [DOI] [PMC free article] [PubMed] [Google Scholar]
78. Keane PA, Topol EJ. With an eye to AI and autonomous diagnosis. NPJ Digit Med. 2018;1:40. 10.1038/s41746-018-0048-y [DOI] [PMC free article] [PubMed] [Google Scholar]
79. Cabitza F, Campagner A, Soares F, et al. The importance of being external. methodological insights for the external validation of machine learning models in medicine. Comput Methods Programs Biomed. 2021;208:106288. 10.1016/j.cmpb.2021.106288 [DOI] [PubMed] [Google Scholar]
80. Bella A, Ferri C, Hernández-Orallo J, Ramírez-Quintana MJ. Calibration of machine learning models. In: Soria Olivas E, Martín Guerrero JD, Martinez Sober M, et al. , eds. Handbook of Research on Machine Learning Applications and Trends: Algorithms, Methods, and Techniques. IGI Global; 2010:128-146. [Google Scholar]
81. Voss EA, Makadia R, Matcho A, et al. Feasibility and utility of applications of the common data model to multiple, disparate observational health databases. J Am Med Inform Assoc. 2015;22:553-564. 10.1093/jamia/ocu023 [DOI] [PMC free article] [PubMed] [Google Scholar]
82. Rosenbloom ST, Carroll RJ, Warner JL, Matheny ME, Denny JC. Representing knowledge consistently across health systems. Yearb Med Inform. 2017;26:139-147. 10.15265/iy-2017-018 [DOI] [PMC free article] [PubMed] [Google Scholar]
83. Garza M, Del Fiol G, Tenenbaum J, Walden A, Zozus MN. Evaluating common data models for use with a longitudinal community registry. J Biomed Inform. 2016;64:333-341. 10.1016/j.jbi.2016.10.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
84. Liu Z, Li X, Peng H, He L, Yu PS. . Heterogeneous Similarity Graph Neural Network on Electronic Health Records. In: 2020 IEEE International Conference on Big Data (Big Data). IEEE Computer Society; 2020:1196-1205. 10.1109/BigData50022.2020.9377795 [DOI]
85. Zhang C, Mohan K, Pearl J. Causal inference with Non-IID data using linear graphical models. Proceedings of the 36th International Conference on Neural Information Processing Systems. Curran Associates Inc., 2024:Article 960.
86. Tang R, Han X, Jiang X, Hu X. Does synthetic data generation of LLMs help clinical text mining? arXiv preprint arXiv:2303.04360. Published March 8, 2023. Accessed May 5, 2024. 10.48550/arXiv.2303.04360 [DOI]
87. Katzman JL, Shaham U, Cloninger A, Bates J, Jiang T, Kluger Y. DeepSurv: personalized treatment recommender system using a Cox proportional hazards deep neural network. BMC Med Res Methodol. 2018;18:24. 10.1186/s12874-018-0482-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
88. Ching T, Zhu X, Garmire LX. Cox-nnet: An artificial neural network method for prognosis prediction of high-throughput omics data. PLoS Comput Biol. 2018;14:e1006076. 10.1371/journal.pcbi.1006076 [DOI] [PMC free article] [PubMed] [Google Scholar]
89. Lee C, Yoon J, Schaar MV. Dynamic-DeepHit: a deep learning approach for dynamic survival analysis with competing risks based on longitudinal data. IEEE Trans Biomed Eng. 2020;67:122-133. 10.1109/tbme.2019.2909027 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

ocae278_Supplementary_Data

ocae278_supplementary_data.docx^{(65.3KB, docx)}

Data Availability Statement

All data generated during this study are provided as supplementary materials.

[ocae278-B1] 1. The Medicare Payment Advisory Commission. Chapter 8. In July 2023 Data Book: Health Care Spending and the Medicare Program. The Medicare Payment Advisory Commission, 2023.

[ocae278-B2] 2. Wang YC, Chou MY, Liang CK, Peng LN, Chen LK, Loh CH. Post-acute care as a key component in a healthcare system for older adults. Ann Geriatr Med Res. 2019;23:54-62. 10.4235/agmr.19.0009 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B3] 3. Centers for Medicare & Medicaid Services. Medicare and Medicaid Programs: Revisions to Requirements for Discharge Planning for Hospitals, Critical Access Hospitals, and Home Health Agencies. Federal Register. Accessed March 9, 2024. https://www.federalregister.gov/documents/2019/09/30/2019-20732/medicare-and-medicaid-programs-revisions-to-requirements-for-discharge-planning-for-hospitals

[ocae278-B4] 4. National Center for Health Statistics. Biennial Overview of Post-Acute and Long-Term Care in the United States. Centers for Disease Control and Prevention. Accessed January 8, 2024. https://data.cdc.gov/d/wibz-pb5q

[ocae278-B5] 5. Valiathan R, Ashman M, Asthana D. Effects of ageing on the immune system: infants to elderly. Scand J Immunol. 2016;83:255-266. 10.1111/sji.12413. [DOI] [PubMed] [Google Scholar]

[ocae278-B6] 6. Gavazzi G, Krause K-H. Ageing and infection. Lancet Infect Dis. 2002;2:659-666. 10.1016/S1473-3099(02)00437-1. [DOI] [PubMed] [Google Scholar]

[ocae278-B7] 7. El Chakhtoura NG, Bonomo RA, Jump RLP. Influence of Aging and Environment on Presentation of Infection in Older Adults. Infect Dis Clin North Am. 2017;31:593-608. 10.1016/j.idc.2017.07.017. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B8] 8. Smith LA, Oakden-Rayner L, Bird A, et al. Machine learning and deep learning predictive models for long-term prognosis in patients with chronic obstructive pulmonary disease: a systematic review and meta-analysis. Lancet Digit Health. 2023;5:e872-e81. 10.1016/s2589-7500(23)00177-2 [DOI] [PubMed] [Google Scholar]

[ocae278-B9] 9. Kline KA, Bowdish DM. Infection in an aging population. Curr Opin Microbiol. 2016;29:63-67. 10.1016/j.mib.2015.11.003 [DOI] [PubMed] [Google Scholar]

[ocae278-B10] 10. Mitchell SL, Teno JM, Kiely DK, et al. The clinical course of advanced dementia. N Engl J Med. 2009;361:1529-1538. 10.1056/NEJMoa0902234 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B11] 11. Givens JL, Selby K, Goldfeld KS, Mitchell SL. Hospital transfers of nursing home residents with advanced dementia. J Am Geriatr Soc. 2012;60:905-909. 10.1111/j.1532-5415.2012.03919.x [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B12] 12. Ouslander JG, Bonner A, Herndon L, Shutes J. The Interventions to Reduce Acute Care Transfers (INTERACT) quality improvement program: an overview for medical directors and primary care clinicians in long term care. J Am Med Dir Assoc. 2014;15:162-170. 10.1016/j.jamda.2013.12.005 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B13] 13. Tingström P, Milberg A, Rodhe N, Ernerud J, Grodzinsky E, Sund-Levander M. Nursing assistants: “he seems to be ill” - a reason for nurses to take action: validation of the Early Detection Scale of Infection (EDIS). BMC Geriatr. 2015;15:122. 10.1186/s12877-015-0114-0 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B14] 14. Vogenberg FR. Predictive and prognostic models: implications for healthcare decision-making in a modern recession. Am Health Drug Benefits. 2009;2:218-222. [PMC free article] [PubMed] [Google Scholar]

[ocae278-B15] 15. Kawamoto K, McDonald CJ. Designing, conducting, and reporting clinical decision support studies: recommendations and call to action. Ann Intern Med. 2020;172:S101-S109. 10.7326/m19-0875 [DOI] [PubMed] [Google Scholar]

[ocae278-B16] 16. Andaur Navarro CL, Damen JAA, van Smeden M, et al. Systematic review identifies the design and methodological conduct of studies on machine learning-based prediction models. J Clin Epidemiol. 2023;154:8-22. 10.1016/j.jclinepi.2022.11.015. [DOI] [PubMed] [Google Scholar]

[ocae278-B17] 17. Theodosiou AA, Read RC. Artificial intelligence, machine learning and deep learning: Potential resources for the infection clinician. J Infect. 2023;87:287-294. [DOI] [PubMed] [Google Scholar]

[ocae278-B18] 18. Jordan MI, Mitchell TM. Machine learning: Trends, perspectives, and prospects. Science. 2015;349:255-260. 10.1126/science.aaa8415 [DOI] [PubMed] [Google Scholar]

[ocae278-B19] 19. Moons KGM, Wolff RF, Riley RD, et al. PROBAST: a tool to assess risk of bias and applicability of prediction model studies: explanation and elaboration. Ann Intern Med. 2019;170:W1-W33. 10.7326/m18-1377 [DOI] [PubMed] [Google Scholar]

[ocae278-B20] 20. Moons KG, de Groot JA, Bouwmeester W, et al. Critical appraisal and data extraction for systematic reviews of prediction modelling studies: the CHARMS checklist. PLoS Med. 2014;11:e1001744. 10.1371/journal.pmed.1001744 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B21] 21. Xu M, Ouyang L, Han L, et al. Accurately differentiating between patients with COVID-19, patients with other viral infections, and healthy individuals: multimodal late fusion learning approach. J Med Internet Res. 2021;23:e25535. 10.2196/25535 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B22] 22. Ren H, Ling Y, Cao R, Wang Z, Li Y, Huang T. Early warning of emerging infectious diseases based on multimodal data. Biosaf Health. 2023;5:229-236. 10.1016/j.bsheal.2023.05.006 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B23] 23. Masot O, Cox A, Mold F, et al. Decision support-tools for early detection of infection in older people (aged> 65 years): a scoping review. BMC Geriatr. 2022;22:552. 10.1186/s12877-022-03218-w [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B24] 24. Yan MY, Gustad LT, Nytrø Ø. Sepsis prediction, early detection, and identification using clinical text for machine learning: a systematic review. J Am Med Inform Assoc. 2022;29:559-575. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B25] 25. Goh KH, Wang L, Yeow AYK, et al. Artificial intelligence in sepsis early prediction and diagnosis using unstructured data in healthcare. Nat Commun. 2021;12:711. 10.1038/s41467-021-20910-4 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B26] 26. Enshaeifar S, Zoha A, Skillman S, et al. Machine learning methods for detecting urinary tract infection and analysing daily living activities in people with dementia. PLoS One. 2019;14:e0209909. 10.1371/journal.pone.0209909 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B27] 27. Stokes K, Castaldo R, Federici C, et al. The use of artificial intelligence systems in diagnosis of pneumonia via signs and symptoms: A systematic review. Biomedical Signal Processing and Control. 2022;72:103325. 10.1016/j.bspc.2021.103325. [DOI] [Google Scholar]

[ocae278-B28] 28. Zhu C, Xu Z, Gu Y, et al. Prediction of post-stroke urinary tract infection risk in immobile patients using machine learning: an observational cohort study. J Hosp Infect. 2022;122:96-107. 10.1016/j.jhin.2022.01.002. [DOI] [PubMed] [Google Scholar]

[ocae278-B29] 29. Fang C, Yan W, Chen Y, et al. Long-term SARS-CoV-2 neutralizing antibody level prediction using multimodal deep learning: A prospective cohort study on longitudinal data in Wuhan, China. J Med Virol. 2023;95:e29036. 10.1002/jmv.29036. [DOI] [PubMed] [Google Scholar]

[ocae278-B30] 30. Adams V, Song J, Shang J, et al. Infection prevention and control practices in the home environment: Examining enablers and barriers to adherence among home health care nurses. Am J Infect Control. 2021;49:721-726. 10.1016/j.ajic.2020.10.021. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B31] 31. Strupp J, Valerius L, Herrler A, et al. I see the high level of interpersonal support during the pandemic as proof that the good in people prevails': the COVID-19 pandemic as a catalyst for building a caring community? J Public Health (Oxf). 2024;46:175-184. 10.1093/pubmed/fdad283 [DOI] [PubMed] [Google Scholar]

[ocae278-B32] 32. Katz MJ, Roghmann MC. Healthcare-associated infections in the elderly: what's new. Curr Opin Infect Dis. 2016;29:388-393. 10.1097/qco.0000000000000283 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B33] 33. Siegel EO, Backman A, Cai Y, et al. Understanding contextual differences in residential LTC provision for cross-national research: identifying internationally relevant CDEs. Gerontol Geriatr Med. 2019;5:2333721419840591. 10.1177/2333721419840591 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B34] 34. Lee DS, Wang CX, McAlister FA, et al. Factors associated with SARS-CoV-2 test positivity in long-term care homes: a population-based cohort analysis using machine learning. The Lancet Regional Health—Americas. 2022;6:100146. 10.1016/j.lana.2021.100146 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B35] 35. Lee DS, Ma S, Chu A, CORONA Collaboration, et al. Predictors of mortality among long-term care residents with SARS-CoV-2 infection. Journal of the American Geriatrics Society. 2021;69:3377-3388. 10.1111/jgs.17425. [DOI] [PubMed] [Google Scholar]

[ocae278-B36] 36. Stone PW, Herzig CTA, Agarwal M, Pogorzelska-Maziarz M, Dick AW. Nursing home infection control program characteristics, CMS citations, and implementation of antibiotic stewardship policies: a national study. Inquiry. 2018;55:46958018778636. 10.1177/0046958018778636 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B37] 37. Houben F, Heijer CDD, Dukers-Muijrers NH, Smeets-Peels C, Hoebe CJ. Psychosocial determinants associated with healthcare workers' self-reported compliance with infection prevention and control during the COVID-19 pandemic: a cross-sectional study in Dutch residential care facilities for people with intellectual and developmental disabilities. BMC Public Health. 2023;23:2052. 10.1186/s12889-023-16912-0 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B38] 38. Kilanowski JF. Breadth of the socio-ecological model. J Agromedicine. 2017;22:295-297. 10.1080/1059924X.2017.1358971 [DOI] [PubMed] [Google Scholar]

[ocae278-B39] 39. Bronfenbrenner U. Ecology of the family as a context for human development: Research perspectives. Developmental Psychology. 1986;22:723-742. 10.1037/0012-1649.22.6.723 [DOI] [Google Scholar]

[ocae278-B40] 40. Mhasawade V, Zhao Y, Chunara R. Machine learning and algorithmic fairness in public and population health. Nat Mach Intell. 2021;3:659-666. [Google Scholar]

[ocae278-B41] 41. Matthew JP, Joanne EM, Patrick MB, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021;372:n71. 10.1136/bmj.n71 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B42] 42. Badillo S, Banfai B, Birzele F, et al. An introduction to machine learning. Clin Pharmacol Ther. 2020;107:871-885. 10.1002/cpt.1796. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B43] 43. Goldstein BA, Navar AM, Carter RE. Moving beyond regression techniques in cardiovascular risk prediction: applying machine learning to address analytic challenges. Eur Heart J. 2017;38:1805-1814. 10.1093/eurheartj/ehw302 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B44] 44. Bramer WM, Giustini D, de Jonge GB, Holland L, Bekhuis T. De-duplication of database search results for systematic reviews in EndNote. J Med Libr Assoc. 2016;104:240-243. 10.3163/1536-5050.104.3.014 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B45] 45. Wolff RF, Moons KGM, Riley RD, PROBAST Group†, et al. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies. Annals of Internal Medicine. 2019;170:51-58. 10.7326/M18-1376 [DOI] [PubMed] [Google Scholar]

[ocae278-B46] 46. Pham MT, Rajić A, Greig JD, Sargeant JM, Papadopoulos A, McEwen SA. A scoping review of scoping reviews: advancing the approach and enhancing the consistency. Res Synth Methods. 2014;5:371-385. 10.1002/jrsm.1123 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B47] 47. Tan PY, Huo M, Zhou XH, Zhao BL. Development and validation of a nomogram for predicting the risk of nursing home-acquired pneumonia. Eur Rev Med Pharmacol Sci. 2022;26:8276-8288. 10.26355/eurrev_202211_30360 [DOI] [PubMed] [Google Scholar]

[ocae278-B48] 48. Baldominos A, Ogul H, Colomo-Palacios R, Sanz-Moreno J, Gómez-Pulido JM. Infection prediction using physiological and social data in social environments. Information Processing & Management. 2020;57:102213. 10.1016/j.ipm.2020.102213. [DOI] [Google Scholar]

[ocae278-B49] 49. Garcés-Jiménez A, Calderón-Gómez H, Gómez-Pulido JM, et al. Medical prognosis of infectious diseases in nursing homes by applying machine learning on clinical data collected in cloud microservices. Int J Environ Res Public Health. 2021;18 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B50] 50. Mehr DR, Binder EF, Kruse RL, et al. Predicting mortality in nursing home residents with lower respiratory tract infection: the Missouri LRI study. JAMA. 2001;286:2427-2436. 10.1001/jama.286.19.2427 [DOI] [PubMed] [Google Scholar]

[ocae278-B51] 51. Naughton BJ, Mylotte JM, Tayara A. Outcome of nursing home-acquired pneumonia: derivation and application of a practical model to predict 30 day mortality. J Am Geriatr Soc. 2000;48:1292-1299. 10.1111/j.1532-5415.2000.tb02604.x. [DOI] [PubMed] [Google Scholar]

[ocae278-B52] 52. Panagiotou OA, Kosar CM, White EM, et al. Risk factors associated with all-cause 30-day mortality in nursing home residents with COVID-19. JAMA Intern Med. 2021;181:439-448. 10.1001/jamainternmed.2020.7968 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B53] 53. Rauh SP, Heymans MW, van der Maaden T, et al. Predicting mortality in nursing home residents with dementia and pneumonia treated with antibiotics: validation of a prediction model in a more recent population. J Gerontol A Biol Sci Med Sci. 2019;74:1922-1928. 10.1093/gerona/gly260 [DOI] [PubMed] [Google Scholar]

[ocae278-B54] 54. van der Steen JT, Mehr DR, Kruse RL, et al. Predictors of mortality for lower respiratory infections in nursing home residents with dementia were validated transnationally. J Clin Epidemiol. 2006;59:970-979. 10.1016/j.jclinepi.2005.12.005. [DOI] [PubMed] [Google Scholar]

[ocae278-B55] 55. Ramazi R, Bowen ME, Flynn AJ, Beheshti R. Developing acute event risk profiles for older adults with dementia in long-term care using motor behavior clusters derived from deep learning. J Am Med Dir Assoc. 2022;23:1977-1983.e1. 10.1016/j.jamda.2022.04.009. [DOI] [PubMed] [Google Scholar]

[ocae278-B56] 56. Shang J, Russell D, Dowding D, et al. A Predictive risk model for infection-related hospitalization among home healthcare patients. J Healthc Qual. 2020;42:136-147. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B57] 57. Song J, Woo K, Shang J, Ojo M, Topaz M. Predictive risk models for wound infection-related hospitalization or ED visits in home health care using machine-learning algorithms. Adv Skin Wound Care. 2021;34:1-12. 10.1097/01.Asw.0000755928.30524.22 [DOI] [PubMed] [Google Scholar]

[ocae278-B58] 58. Maritz R, Ehrmann C, Prodinger B, Tennant A, Stucki G. The influence and added value of a Standardized Assessment and Reporting System for functioning outcomes upon national rehabilitation quality reports. Int J Qual Health Care. 2020;32:379-387. 10.1093/intqhc/mzaa058 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B59] 59. Mohsen F, Ali H, El Hajj N, Shah Z. Artificial intelligence-based methods for fusion of electronic health records and imaging data. Sci Rep. 2022;12:17981. 10.1038/s41598-022-22514-4 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B60] 60. Huang S-C, Pareek A, Seyyedi S, Banerjee I, Lungren MP. Fusion of medical imaging and electronic health records using deep learning: a systematic review and implementation guidelines. NPJ Digit Med. 2020;3:136. 10.1038/s41746-020-00341-z [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B61] 61. Calders T, Kamiran F, Pechenizkiy M. Building classifiers with independency constraints. In: 2009 IEEE International Conference on Data Mining Workshops, Miami, FL, USA. IEEE Computer Society; 2009:13-18.. 10.1109/ICDMW.2009.83 [DOI]

[ocae278-B62] 62. Hardt M, Price E, Srebro N. Equality of opportunity in supervised learning. Proceedings of the 30th International Conference on Neural Information Processing Systems. Curran Associates Inc., 2016:3323-31.

[ocae278-B63] 63. Harrison JM, Dick AW, Stone PW, et al. Infection trends in home health care, 2013–2018. Infect Control Hosp Epidemiol. 2021;42:1388-1390. 10.1017/ice.2021.248 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B64] 64. Naegelin M, Weibel RP, Kerr JI, et al. An interpretable machine learning approach to multimodal stress detection in a simulated office environment. J Biomed Inform. 2023;139:104299. 10.1016/j.jbi.2023.104299. [DOI] [PubMed] [Google Scholar]

[ocae278-B65] 65. Coudray N, Ocampo PS, Sakellaropoulos T, et al. Classification and mutation prediction from non–small cell lung cancer histopathology images using deep learning. Nat Med. 2018;24:1559-1567. 10.1038/s41591-018-0177-5 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B66] 66. Association AH. Fact Sheet: Post-acute Care. Secondary Fact Sheet: Post-acute Care 2019. https://www.aha.org/system/files/media/file/2019/07/fact-sheet-post-acute-care-0719.pdf.

[ocae278-B67] 67. Pinkas G, Karny Y, Malachi A, Barkai G, Bachar G, Aharonson V. SARS-CoV-2 detection from voice. IEEE Open J Eng Med Biol. 2020;1:268-274. 10.1109/ojemb.2020.3026468 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B68] 68. Xiong Z, Achavananthadith S, Lian S, et al. A wireless and battery-free wound infection sensor based on DNA hydrogel. Sci Adv. 2021;7:eabj1617. 10.1126/sciadv.abj1617 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B69] 69. Centers for Medicare & Medicaid Services. The CMS Framework for Health Equity 2022-2032. 2022. https://www.cms.gov/files/document/cms-framework-health-equity-2022.pdf

[ocae278-B70] 70. WHO Regional Office for South-East Asia. Social Determinants of Health. 2008

[ocae278-B71] 71. Yearby R. Structural racism and health disparities: reconfiguring the social determinants of health framework to include the root cause. J Law Med Ethics. 2020;48:518-526. [DOI] [PubMed] [Google Scholar]

[ocae278-B72] 72. Patra BG, Sharma MM, Vekaria V, et al. Extracting social determinants of health from electronic health records using natural language processing: a systematic review. J Am Med Inform Assoc. 2021;28:2716-2727. 10.1093/jamia/ocab170 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B73] 73. Rehman N, Saif U, Chunara R. Deep landscape features for improving vector-borne disease prediction. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops. IEEE; 2019:44-51.

[ocae278-B74] 74. Kline A, Wang H, Li Y, et al. Multimodal machine learning in precision health: A scoping review. NPJ Digit Med. 2022;5:171. 10.1038/s41746-022-00712-8 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B75] 75. Van Calster B, Wynants L, Timmerman D, Steyerberg EW, Collins GS. Predictive analytics in health care: how can we know it works? J Am Med Inform Assoc. 2019;26:1651-1654. 10.1093/jamia/ocz130 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B76] 76. Lundberg SM, Lee S-I. A unified approach to interpreting model predictions. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. Curran Associates Inc.; 2017:4768-4777. [Google Scholar]

[ocae278-B77] 77. Payrovnaziri SN, Chen Z, Rengifo-Moreno P, et al. Explainable artificial intelligence models using real-world electronic health record data: a systematic scoping review. J Am Med Inform Assoc. 2020;27:1173-1185. 10.1093/jamia/ocaa053 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B78] 78. Keane PA, Topol EJ. With an eye to AI and autonomous diagnosis. NPJ Digit Med. 2018;1:40. 10.1038/s41746-018-0048-y [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B79] 79. Cabitza F, Campagner A, Soares F, et al. The importance of being external. methodological insights for the external validation of machine learning models in medicine. Comput Methods Programs Biomed. 2021;208:106288. 10.1016/j.cmpb.2021.106288 [DOI] [PubMed] [Google Scholar]

[ocae278-B80] 80. Bella A, Ferri C, Hernández-Orallo J, Ramírez-Quintana MJ. Calibration of machine learning models. In: Soria Olivas E, Martín Guerrero JD, Martinez Sober M, et al. , eds. Handbook of Research on Machine Learning Applications and Trends: Algorithms, Methods, and Techniques. IGI Global; 2010:128-146. [Google Scholar]

[ocae278-B81] 81. Voss EA, Makadia R, Matcho A, et al. Feasibility and utility of applications of the common data model to multiple, disparate observational health databases. J Am Med Inform Assoc. 2015;22:553-564. 10.1093/jamia/ocu023 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B82] 82. Rosenbloom ST, Carroll RJ, Warner JL, Matheny ME, Denny JC. Representing knowledge consistently across health systems. Yearb Med Inform. 2017;26:139-147. 10.15265/iy-2017-018 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B83] 83. Garza M, Del Fiol G, Tenenbaum J, Walden A, Zozus MN. Evaluating common data models for use with a longitudinal community registry. J Biomed Inform. 2016;64:333-341. 10.1016/j.jbi.2016.10.016. [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B84] 84. Liu Z, Li X, Peng H, He L, Yu PS. . Heterogeneous Similarity Graph Neural Network on Electronic Health Records. In: 2020 IEEE International Conference on Big Data (Big Data). IEEE Computer Society; 2020:1196-1205. 10.1109/BigData50022.2020.9377795 [DOI]

[ocae278-B85] 85. Zhang C, Mohan K, Pearl J. Causal inference with Non-IID data using linear graphical models. Proceedings of the 36th International Conference on Neural Information Processing Systems. Curran Associates Inc., 2024:Article 960.

[ocae278-B86] 86. Tang R, Han X, Jiang X, Hu X. Does synthetic data generation of LLMs help clinical text mining? arXiv preprint arXiv:2303.04360. Published March 8, 2023. Accessed May 5, 2024. 10.48550/arXiv.2303.04360 [DOI]

[ocae278-B87] 87. Katzman JL, Shaham U, Cloninger A, Bates J, Jiang T, Kluger Y. DeepSurv: personalized treatment recommender system using a Cox proportional hazards deep neural network. BMC Med Res Methodol. 2018;18:24. 10.1186/s12874-018-0482-1 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B88] 88. Ching T, Zhu X, Garmire LX. Cox-nnet: An artificial neural network method for prognosis prediction of high-throughput omics data. PLoS Comput Biol. 2018;14:e1006076. 10.1371/journal.pcbi.1006076 [DOI] [PMC free article] [PubMed] [Google Scholar]

[ocae278-B89] 89. Lee C, Yoon J, Schaar MV. Dynamic-DeepHit: a deep learning approach for dynamic survival analysis with competing risks based on longitudinal data. IEEE Trans Biomed Eng. 2020;67:122-133. 10.1109/tbme.2019.2909027 [DOI] [PubMed] [Google Scholar]

PERMALINK

Machine learning-based infection diagnostic and prognostic models in post-acute care settings: a systematic review

Zidu Xu, MPhil, MMed

Danielle Scharp, PhD, APRN

Mollie Hobensack, PhD, RN

Jiancheng Ye, PhD

Jungang Zou, MS

Sirui Ding, PhD

Jingjing Shang, PhD, RN, FAAN

Maxim Topaz, PhD, MA, RN, FAAN, FIAHSI, FACMI

Abstract

Objectives

Materials and Methods

Results

Discussion and Conclusion

Introduction

Figure 1.

Objectives

Methods

Registration

Eligibility criteria

Information sources and search strategy

Study selection and data extraction

Quality appraisal

Data synthesis and conceptual framework

Figure 2.

Results

Study selection

Study characteristics

Table 1.

Study quality evaluation

Figure 3.

Data usage

Risk predictors

Figure 4.

ML algorithms

Validation techniques and evaluation metrics

Model performances

Table 2.

Model interpretability

Model translation and clinical applications

Discussion

Figure 5.

Limitations

Conclusions

Supplementary Material

Acknowledgments

Contributor Information

Supplementary material

Author contributions

Funding

Conflicts of interest

Data availability

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases