Factors associated with poor self-management documented in home health care narrative notes for patients with heart failure

Abstract

Background:

Patients with heart failure (HF) who actively engage in their own self-management have better outcomes. Extracting data through natural language processing (NLP) holds great promise for identifying patients with or at risk of poor self-management.

Objective:

To identify home health care (HHC) patients with HF who have poor self-management using NLP of narrative notes, and to examine patient factors associated with poor self-management.

Methods:

An NLP algorithm was applied to extract poor self-management documentation using 353,718 HHC narrative notes of 9,710 patients with HF. Sociodemographic and structured clinical data were incorporated into multivariate logistic regression models to identify factors associated with poor self-management.

Results:

There were 758 (7.8%) patients in this sample identified as having notes with language describing poor HF self-management. Younger age (OR 0.982, 95% CI 0.976–0.987, p < .001), longer length of stay in HHC (OR 1.036, 95% CI 1.029– 1.043, p < .001), diagnosis of diabetes (OR 1.47, 95% CI 1.3–1.67, p < .001) and depression (OR 1.36, 95% CI 1.09–1.68, p < .01), impaired decision-making (OR 1.64, 95% CI 1.37–1.95, p < .001), smoking (OR 1.7, 95% CI 1.4–2.04, p < .001), and shortness of breath with exertion (OR 1.25, 95% CI 1.1–1.42, p < .01) were associated with poor self-management.

Conclusions:

Patients with HF who have poor self-management can be identified from the narrative notes in HHC using novel NLP methods. Meaningful information about the self-management of patients with HF can support HHC clinicians in developing individualized care plans to improve self-management and clinical outcomes.

Introduction

According to the American Heart Association, approximately 6 million adults had heart failure (HF) in 2018, and an estimated 359,119 people died of HF in 2017 in the United States.¹ The estimated cost to treat patients with HF was $43.6 billion in 2020.² In the US, approximately 200,000 patients with HF are discharged from hospitals to home healthcare (HHC) agencies every year.³ The aging population and increasing burden of HF are significant challenges for the healthcare system.⁴ HF requires lifelong care with optimal self-management and effective patient education, which can be supported in the HHC setting.⁵ Self-care (self-management) is defined as the ability to monitor one’s conditions, managing changes when they occur, and achieving the cognitive, behavioral, and emotional actions necessary to maintain a satisfactory quality of life.⁶ Research shows that people who actively engage in chronic disease self-management have better health outcomes and are more likely to appropriately use health care services.⁵ It is important for HHC clinicians to recognize a patient’s self-management status to prevent symptom exacerbations that could result in avoidable complications. Identifying factors associated with patients’ poor self-management might help HCC clinicians plan and implement self-management interventions tailored to individual patients.⁷

Despite the extensive documentation of the Outcome and Assessment Information Set (OASIS data) required of clinicians in HHC,⁸ novel methods are needed to monitor patients’ self-management status using electronic health records (EHRs) so that clinicians can provide timely care and education. Prior studies rely mainly on survey methodology (e.g., Self-Care of Heart Failure Index⁹) to examine self-management in patients with HF^10–13; however, survey methods are costly, labor intensive, and prone to selection and recall bias.

Secondary data from the EHR may help address some of these challenges. However, using EHR data to ascertain self-management status has its own challenges. EHRs often do not have a dedicated, structured field to document self-management status. Most of the relevant documentation can be found in thousands of narrative clinical notes documented for each patient over the course of their care (e.g., “patient is noncompliant”, “patient still has high sodium diet”).¹⁴ Extraction of these data using natural language processing (NLP) holds promise for identifying information to guide optimal self-management. NLP is a collection of computational algorithms that automatically extract, process, and analyze information from written resources to discover clinically significant factors.^15,16 Previous studies have successfully applied NLP to identify patients with clinical risk factors,^16–21 including falls,¹⁹ wounds,²⁰ alcohol and substance abuse status,²¹ adverse events,²² or hospital utilization,^23,24 and to improve risk prediction models.^24,25 NLP methods are currently being used to extract symptom information from a large amount of data in free-text narrative notes in EHRs.^18,26–29 A previous study developed an NLP algorithm to identify HF patients with poor self-management status from narrative discharge summary notes, but in a hospital setting.¹⁴ Our group has previously developed an NLP algorithm to automatically identify HF patients with poor self-management in the HHC setting³⁰ and in the current study we apply this algorithm to a large sample of HF patients receiving HHC services.

In the present study, we (1) identify HHC patients with HF who have documented poor self-management using NLP, and (2) identify factors associated with poor self-management. We used the Self- and Family Management Framework³¹ to guide our selection of specific domains of self-management. The Self- and Family Management Framework comprehensively delineates self-management processes that guide nursing care and can more clearly address how self-management interventions work and under what conditions.³¹ The most recent version of this framework depicts the relationships among the four dimensions: facilitators and barriers, processes, proximal outcomes, and distal outcomes of self- and family management.³¹ We focused on three categories of self-management processes – illness needs, activating resources, and living with a chronic illness – to define the domains of HF self-management. Poor self-management is defined in this study as deficits that patients with HF have in achieving the physical, cognitive, behavioral, and emotional actions necessary to maintain a satisfactory quality of life.⁶ We operationally defined ‘Poor self-management’ as patients with HF who have had at least one instance of documented poor self-management, under the domains of poor diet adherence, medication adherence, or exercise/physical activity adherence; issues with other self-management activities/self-monitoring; missed healthcare encounters; or unspecified non-adherence.

Methods

Study design

A retrospective, observational study with secondary data analysis was conducted using HHC narrative notes from one of the largest HHC agencies in the Northeastern US. Narrative notes were completed by HHC clinicians (registered nurses, social workers, physical and occupational therapists, and clinical team managers) and stored in the HHC agency’s EHR system. Patient sociodemographic and structured clinical assessment data (i.e., OASIS) were also extracted from the HHC agency’s EHR system.

Study population

The sample consisted of 9710 HF patients, with 15,948 HHC Episodes of Care (EOC). A HHC episode was defined as all services provided within a specific period of time (i.e., initiated by a HHC admission, ended by a HHC discharge).

Included in this study were all patients admitted to HHC between January 1, 2015 and December 31, 2017 after being discharged from a hospital with a diagnosis of HF (International Classification of Diseases codes, Tenth Revision, Clinical Modification [ICD-10] codes 50. x, I11.0, I13.0, I13.1, I13.2).³²

Narrative notes

All HHC narrative notes available in the EHR were extracted for this patient cohort. Two types of narrative notes were included: (1) visit notes (total n = 153,326, average note length = 272.3 characters), and (2) care coordination notes (total n = 00,579, average note length = 106.6 characters). Visit notes were generated by clinicians to describe the care provided and the patient’s status during a HHC visit. Care coordination notes document communication between clinicians (e.g., calling a physician) and other care-related activities (e.g., ordering wound care supplies). Duplicate notes were removed (n = 187), resulting in a total of 353,718 narrative notes used in this study.

Study variables

Potential variables associated with poor-self management status were identified based on the study’s conceptual framework and relevant literature.^{7,13,31,33,34}

Selected demographic and clinical characteristics were taken from the OASIS assessments, which are required by the Centers for Medicare and Medicaid. These assessments collect information on nearly 100 items related to a home care recipient’s demographic information, clinical status, functional status, and service needs.³⁵ The variables included gender, age, co-morbid conditions, risk factors such as smoking, behavioral issues, shortness of breath, and the ability to prepare meals and manage medication. Specifically, activities of daily living (ADLs)/instrumental activities of daily living (IADLs) measured the patient’s current ability for grooming, dressing of the upper body, dressing of the lower body, bathing, toilet transferring, toilet hygiene, and transferring. Prior functioning ADLs/IADLs indicated the patient’s ability to perform everyday activities prior to his or her most recent illness, exacerbation, or injury.

Comorbid conditions included arthritis, diabetes, depression, neurological diseases, cardiac dysrhythmias, pulmonary diseases, renal diseases, and skin ulcer. Conditions present prior to the start of the HHC included urinary incontinence, indwelling/suprapubic catheter, intractable, impaired decision-making, disruptive or socially inappropriate behavior, and memory loss. A complete list of study variables is presented in Supplemental Table 1.

Study procedures

NLP algorithm development, validation, and application

Based on the conceptual framework, six domains of HF self-management were identified: poor diet adherence, poor medication adherence, poor exercise/physical activity tolerance, issues with other self-care activities/self-monitoring, missed healthcare encounters, and unspecified non-adherence (i.e., documentation of non-adherence without mention of a specific domain, for example, “Patient remains non-adherent.”).^{7,13,33,34,36,37} Lists of poor self-management categories were further developed in our previous study¹⁴ based on: (a) relevant literature and standard health terminologies (e.g., the Systemized Nomenclature of Medicine – Clinical Terms [SNOMED-CT], the International Classification for Nursing Practice [ICNP^®]); (b) multidisciplinary team’s clinical expertise in nursing, medicine, and pharmacy; and (c) full-text review of a random sample of 200 narrative notes performed by two health informaticians with HHC clinical background.

Next, we used a previously validated, open-source NLP tool called NimbleMiner (https://github.com/mtopaz/NimbleMiner).³⁸ The high accuracy of NimbleMiner’s concept identification using narrative notes in EHRs to extract symptoms and important patient information has been previously reported.^{19–21,26,38} The overall accuracy to identify the concept of poor self-management was high, with precision (also known as positive predictive value; $\frac{\text{true}\text{positives}}{\text{true}\text{positives}+\text{false}\text{positives}})=0.94$ (Supplemental Table 2). The process of developing NLP for identifying poor self-management in HHC was more fully described in a recent publication.³⁰ We applied our previously developed and validated NLP algorithm to the dataset of narrative notes for 9710 patients with HF to divide the episodes of care into two groups: episodes of care with poor self-management and episodes of care without poor self-management. If a patient had at least one instance of poor self-management documented, they were classified as having poor self-management.

Statistical analysis

Descriptive statistics were calculated for the demographic and clinical characteristics of the 9710 patients with HF (15,948 episodes). Univariate analyses (i.e., T-test and Chi-square test) were used to compare the clinical and sociodemographic characteristics between patients with and without documented poor self-management.

Statistically significant (p-value < .05) and clinically meaningful variables were incorporated in a multivariate model to identify the associations between clinical and demographic characteristics, and poor self-management. Multivariate logistic regression using stepwise variable selection based on the Akaike information criterion (AIC) was applied. The AIC for a given model is a function of its log-likelihood and the number of estimable parameters, therefore, we selected the model with the smallest AIC.^39,40 Odds ratios (ORs) were calculated with 95% confidence intervals (CIs). All analyses were implemented using R (version 4.0.3).⁴¹

Results

Patient characteristics

A total of 353,718 notes were identified and analyzed for the 9710 patients included in this analysis. The average number of episodes per patient was 1.64, and the average number of notes per patients was 36.42. In general, this sample of patients with HF were older adults (average age, 81.6 years) and male (60.9%). The most common comorbid conditions were hypertension, diabetes, cardiac dysrhythmias, and acute myocardial infarction with ischemic heart disease (72.3%, 38.8%, 32.6%, and 30.7%, respectively). The average length of stay in HHC was 46 days (SD 56), with a 25% Quartile of 26 and a 75% Quartile of 54.

NLP poor self-management findings

When applying the NLP algorithm, 758/9,710 (7.8%) patients and 1178/15,948 (7.3%) episodes in this sample were identified as having notes with language describing HF poor self-management. Table 1 presents examples of language that were used to describe patients with HF poor self-management. Documentation of unspecified poor HF self-management (e.g., “noncompliance,” “uncontrolled”) was the most prevalent (4.7% of notes, 6% of patients), followed by the domain of poor diet adherence (e.g., “eat a lot, salty food”) (0.7% of notes, 2% of patients). Rarely documented poor HF self-management domains (<0.001% of notes) were Issues with other self-care activities (e.g., “not check bp,” 75/36,811 notes), Poor medication adherence (e.g., “medication noncompliance,” 22/36,811 notes), Missed healthcare encounters (e.g., “did not follow up,” 19/36,811 notes), and Poor exercise and physical activity (e.g., “not exercise,” 2/36,811 notes).”

Table 1.

Examples of narrative notes with poor self-management.

Example notes #1	pt is a 82yo male recently hospitalized for upper lower extremity weakness and swelling after failure of taking methotrexate pt reports feeling generally tired participated in short distance ambulation indoor and *limited exercise activities (under domain of “Poor exercise physical activity adherence”)
Example notes #2	pt continues to admit *noncompliance (under domain of “Unspecified nonadherence”) with exercises and reports *not drinking enough (under domain of “Poor diet adherence”)
Example notes #3	pt *didnt follow up (under domain of “Missed healthcare encounters”) with md for inr and she admitted to binge eating processes popcorn consequences of *high sodium diet (under domain of “Poor diet adherence”) in relation to heart disease

^*Positively labeled terms.

Structured factors associated with poor self-management

All of the analyses were performed at the episode level except demographic characteristics (e.g., age, race, and gender) which were analyzed at the patient level.

Univariate results

Univariate comparisons of the characteristics of patients with and without documented poor HF self-management are summarized in Table 2. In the univariate comparisons, the following 24 factors were found to be significantly associated with poor self-management (Table 2).

Table 2.

The comparison of clinical and demographic characteristics between patients with and without poor self-management.

Clinical demographic profile	Total (n = 9,710,15,948 HHC episodes)	No Poor Self-Management (n = 8,951,14,770 episodes)	PoorSelf-Management (n = 758, 1178 episodes)	p-value
	Patient N(%)a
Age at start of care (mean, years, SD)	81.56 (10.81)	81.78 (10.7)	78.99 (11.9)	<.001
Race
Asian/Others/Unknown /Native Hawaiian or Pacific Islander	3652 (37.62)	3397 (37.94)	255 (33.64)	Ref
non-Hispanic Black	1313 (13.52)	1178 (13.16)	135 (17.81)	<.001
non-Hispanic Caucasian	3622 (37.30)	3337 (37.28)	285 (37.60)	0.158
Hispanic	1123 (11.56)	1040 (11.62)	83 (10.95)	0.655
Gender
Male	3792 (39.05)	3461 (38.66)	331 (43.67)	<.001
	Episode N (%)b
Length ofstay in HHC (mean, days, SD)	46.07 (55.81)	44.12 (44.1)	69.06 (69.1)	<.001
Comorbidities (Mark all that apply)
Arthritis	2221 (13.93)	2092 (14.16)	129 (10.95)	<.001
Diabetes	6183 (38.79)	5596 (37.89)	587 (49.83)	<.001
Depression	1054 (6.61)	953 (6.45)	101 (8.57)	.004
Neurological diseases	516 (3.23)	490 (3.32)	26 (2.21)	0.03
Cardiac dysrhythmias	5195 (32.57)	4849 (32.83)	346 (29.37)	0.01
Pulmonary diseases	4069 (25.51)	3738 (25.31)	331 (28.1)	0.03
Renal diseases	880 (5.51)	800 (5.42)	80 (6.79)	0.04
Skin ulcer	2335 (14.64)	2132 (14.43)	203 (17.23)	0.008
Prior conditions within past 14 days
Disruptive or socially inappropriate behavior	122 (0.76)	102(0.69)	20 (1.7)	<.001
Impaired decision-making	1667 (10.45)	1492 (10.1)	175 (14.86)	<.001
ADL/IADLs: Grooming
Able	1450 (9.09)	1323 (8.96)	127 (10.78)	Ref
Assistance needed	12817 (80.37)	11873 (80.38)	944 (80.14)	0.05
Dependent	1681 (10.54)	1574 (10.66)	107 (9.08)	0.01
ADL/IADLs: Dress Upper
Able	452 (2.83)	405 (2.74)	47 (3.99)	Ref
Assistance needed	13482 (84.54)	12488 (84.55)	994 (84.38)	0.01
Dependent	2014 (12.63)	1877 (12.71)	137 (11.63)	0.008
ADL/IADLs: ToiletTransfer
Able	4009 (25.14)	3675 (24.88)	334 (28.35)	Ref
Assistance needed	10141 (63.59)	9407 (63.69)	734 (62.31)	0.02
Dependent	1798 (11.27)	1688 (11.43)	110 (9.34)	0.002
ADL/IADLs: Transfer
Independent	4683 (29.37)	4302 (29.13)	381 (32.34)	Ref
Needed Some Help	9374 (58.78)	8698 (58.89)	676 (57.39)	0.04
Dependent	1891 (11.85)	1770 (11.98)	121 (10.27)	0.01
Risk Factors (Mark all that apply)
Alcohol dependency	171 (1.07)	149 (1.01)	22 (1.87)	0.005
Drug	96 (0.60)	81 (0.55)	15 (1.27)	0.001
Obesity	2974 (18.64)	2690 (18.21)	284 (24.11)	<.001
Smoking	1170 (7.33)	1023 (6.93)	147 (12.48)	<.001
Shortness of breath
Never	5664 (35.51)	5304 (35.91)	360 (30.56)	Ref
With exertion	9930 (62.28)	9140 (61.88)	790 (67.06)	<.001
At rest	354 (2.21)	326 (2.21)	28 (2.38)	0.27
Current ability to plan and prepare meals
Able	759 (4.76)	691 (4.68)	68 (5.77)	Ref
Not regular Basis	6078 (38.12)	5588 (37.83)	490 (41.60)	0.37
Unable	9111 (57.12)	8491 (57.49)	620 (52.63)	0.03
Management for oral medication
Able	3404 (21.35)	3115 (21.09)	289 (24.53)	Ref
Assistance needed	6566 (41.17)	6076 (41.14)	490 (41.06)	0.06
Unable	5870 (36.81)	5474 (37.06)	396 (33.62)	0.001
No Meds	108 (0.67)	105 (0.71)	3 (0.25)	0.02

Ref = Reference

^aThe comparisons of demographic characteristics were analyzed at the patient level.

^bThe comparisons of clinical characteristics were analyzed at the patient level.

Specifically, HF patients with poor self-management had a higher proportion of male patients compared to those without poor self-management (43.7% vs. 38.7%, p < .001). HF patients with poor self-management had a higher proportion of non-Hispanic Black patients compared to those without poor self-management (17.81% vs. 13.16%, p < .001).

Patients with documented poor self-management had longer length of stay in HHC, compared to those without poor self-management documentation (69 days vs. 44 days, p < .001). Patients with documented poor HF self-management were less likely to be dependent in ADLs/ IADLs related to grooming (9.1% vs. 10.7%, p < .05), dressing the upper body (11.6% vs. 12.7%, p < .01), toilet transfer (9.3% vs. 10.6%, p < .01), and transferring (10.3% vs. 11.2%, p < .05) compared to those without poor self-management documentation.

Patients with documented poor HF self-management had a higher proportion of alcohol dependency, drug use, and obesity (1.87% vs. 1.01%, p = .005, 1.27% vs. 0.55%, p = .001, 24.11% vs. 18.21%, p < .001, respectively), compared to those without poor self-management documentation.

Patients with documented poor HF self-management were more likely to have documented conditions present prior to the start of the HHC, such as impaired decision-making (14.8% vs. 10.1%, p < .001), and disruptive or socially inappropriate behavior (1.7% vs. 0.7%, p < .001), compared to those without poor self-management documentation. Finally, patients with poor self-management documentation were more likely to have diabetes (49.8% vs. 37.9%, p < .001), depression (8.6% vs. 6.6%, p < .001), and any type of skin ulcer (17.3% vs. 14.4%, p < .001), compared to those without poor self-management documentation.

Factors associated with poor self-management from multivariate model

To analyze factors that might influence poor self-management, the following variables were incorporated in multivariate regression model: age, race, gender, length of stay, comorbidities (arthritis, diabetes, depression, neurological diseases, cardiac arrhythmias, pulmonary diseases, renal diseases, skin ulcer), conditions present prior to the start of the HHC (disruptive or socially inappropriate behavior, impaired decision-making), presence of ADL limitations (grooming, upper body dressing, toilet transfer, transferring), presence of risk factors (drug dependency, obesity, smoking), shortness of breath with exertion, ability to have meals, and ability to manage oral medication (all p < .05).

The results of multivariate analysis for the associations between the clinical and sociodemographic characteristics and documentation of poor self-management are represented in Table 3. In multivariate analysis, patients with documented poor self-management were younger (OR 0.982 for each 1 year increment, 95% CI 0.976–0.987, p < .001), more likely to be male (OR 1.15, 95% CI 1.01–1.30, p < .05), to have longer HHC lengths of stay (OR 1.036 for each10 day increase, 95% CI 1.029–1.043, p < .001), to have diabetes (OR 1.47, 95% CI 1.30–1.67, p < .001) and depression (OR 1.36, 95% CI 1.09–1.68, p < .01), to exhibit disruptive or socially inappropriate behavior prior to the start of HHC (OR 2.05, 95% CI 1.20–3.33, p < .01), to have impaired decision-making within 14 days prior to start of HHC (OR 1.64, 95% CI 1.37–1.95, p < .001), to be a smoker (OR 1.70, 95% CI 1.40–2.04, p < .001), and to have shortness of breath with exertion (OR 1.25, 95% CI 1.10–1.43, p < .01).

Table 3.

Factors associated with Poor Self-Management

Variables	Multivariate Analysis Adjusted OR (95% CI)
Age (per 1 year increase)	0.982 (0.976–0.987) ***
Male gender (Ref = female)	1.15(1.01–1.30) *
Length of stay in HHC (per 10 days increase)	1.036(1.029–1.043) ***
Diagnosis with any diabetes (Ref = no)	1.47(1.30–1.67) ***
Diagnosis with any depression (Ref = no)	1.36(1.09–1.68) **
Disruptive or socially inappropriate	2.05(1.20–3.33) **
behavior observed within 14 days prior to start ofHHC (Ref = no)
Impaired decision-making observed within 14 days prior to start of HHC (Ref = no)	1.64(1.37–1.95) ***
Risk factors - smoking (Ref= no)	1.70(1.40–2.04) ***
Dyspnea	Ref
Notshort ofbreath
Short of breath with exertion	1.25(1.10–1.43)**

Ref = Reference

^***p < .001

^**p < .01

^*p < .05

Discussion

This study demonstrated that narrative notes in HHC can be extracted from the EHR and analyzed using novel NLP methods. This study also uncovered meaningful information about the factors associated with documented poor self-management in HF patients. The results are consistent with previous literature that younger age,^5,42 less experience with disease,³³ and comorbid conditions such as depression,^{12,33,43–47} are associated with poor self-management.

Fewer than one out of every ten patients (7.8%) with HF in this sample had documented poor self-management. This is lower than a previous study that found approximately 14% of HF patients had an indication of poor self-management in a hospital setting.¹⁴ As observed in the previous study, documentation of unspecified poor HF self-management was most common (8.6% of patients), followed by poor diet adherence (5.2%). Our findings show similar documentation trends, where we likewise found unspecified poor self-management and poor diet adherence to be the most commonly documented domains.

The overall difference in the amount of documented poor self-management in narrative notes between HHC and hospitals suggests that patients might be more engaged in self-management and have increased self-efficacy in HHC than in hospital settings. Also, patients in acute care settings might be discouraged from active self-management, as they are often confined to bed and are dependent on the hospital staff for medications, meals, and bathing. Another hypothesis is that HHC nurses are able to directly observe patients performing self-management activities whereas hospital staff may not witness the same level of patient self-management. During health transitions from hospital to home, patients need to develop self-management skills that will allow them to reduce negative outcomes.⁴⁸ Identifying individual factors associated with poor self-management across this transition may help us to better understand the challenges faced by patients and provide opportunities for supporting patient’s self-management needs.

The final set of factors associated with poor self-management identified in the multivariate regression analysis included the following eight variables: male gender, longer length of stay in HHC, diagnosis of diabetes, diagnosis of depression, disruptive or socially inappropriate behavior and impaired decision-making prior to the start of HHC, smoking, and shortness of breath with exertion. Our finding that poor self-management is less likely to be documented in older patients is supported by prior research showing that increased age contributes to better HF self-management.⁴² Previous studies have shown that patients who have lived with HF for longer period time gain knowledge and are better able to perform daily self-care.^33,43

Male patients with HF had a 15% higher odds of documented poor self-management compared to female patients with HF. There are few studies that report gender differences in terms of compliance with self-care behavior; however, male gender was a significant predictor of poor self-care maintenance in several previous studies.^49–52 Male gender may be associated with having a higher-sodium diet and lower overall dietary quality.⁵³ In addition, physical and social restrictions affecting daily life tend to be the most burdensome for men while limitations that affect the ability to support family and friends tend to be the most challenging to accept for women.⁵² Factors such as these may influence the observed gender differences in having documented poor self-management.

Poor self-management was more likely to be documented as the length of stay in HHC increased. It could be that the longer length of stay allowed for greater opportunity to document poor self-management, or it could be that patients with HF who have a longer length of stay have greater need for care planning and education. Accordingly, self-management programs may wish to target patients with prolonged length of stay in HHC.

Several comorbid conditions were associated with having documented poor self-management. HF patients who were diagnosed with diabetes were more likely to have documented poor HF self-management. Patients with comorbidities, such as diabetes, often take a higher number of medications and are more likely experience discomfort,⁵⁴ which may make self-management more difficult or burdensome. Similarly, patients with depression were more likely to have documented poor-self management in their narrative notes, consistent with other studies in which depression is associated with poor adherence to medical regimens and self-care.^{12,44–47,55} Furthermore, patients who had disruptive or socially inappropriate behaviors and impaired decision-making were more likely to have documented poor self-management compared to patients without these behavioral conditions. This is also consistent with previous literature that suggests that self-management is an iterative process of decisions and behaviors, whereby impaired decision-making may diminish the patient’s ability to self-manage complex chronic conditions.^56,57 This is especially relevant in HHC since patients with depression are more likely to suffer from disruptive behaviors and impaired decision-making.⁵⁸ HHC clinicians may need to prioritize patients with such conditions for targeted HF self-management educational interventions.

Our results show that patients who are smokers often have poor self-management. Self-management in HF involves behavioral adaptation such as monitoring symptoms and managing complex medical regimens, as well as smoking cessation, restricting sodium, and regular exercise.⁵⁹ HHC clinicians can help patients experience a sense of control and achievement for abstaining from smoking⁵⁹ and for performing other self-management behaviors.

Patients identified as having shortness of breath with exertion had higher odds of documented poor-self management in their narrative notes. Shortness of breath is a symptom of HF that contributes to poor self-management because shortness of breath and physical limitations may impact patients’ ability to care for themselves. Our findings suggest that clinicians should more carefully assess patient’s shortness of breath, how it impacts on ability to perform daily living, and consider coping strategies to accommodate this debilitating symptom that could improve their self-management abilities.⁶⁰

Tailored self-management education is an important tool for helping patients with HF. However, self-management information is not consistently recorded in the EHR. We attempted to overcome this limitation of the EHR system by employing NLP to extract useful information about patient self-management contained within the narrative notes. Importantly, this is the first study to use NLP approaches to identify self-management status for HHC patients with HF. This study can serve as a prototype for accessing free text EHR data through NLP to uncover areas of poor self-management in HF patients. The ability to extract self-management information from EHR notes can advance the science of self-management by providing access to rich data that has often been neglected.

Recent studies show that health professionals report a high clinical documentation burden, with up to 50% of clinical time being spent on documenting in EHRs and reading reports documented by other clinicians rather than direct clinical care.⁶¹ NLP techniques applied in this study can be used to develop computerized decision support systems to make use of this rich data to aid clinicians in identifying patients with documented poor self-management. This might save time and reduce clinical documentation review burden.

In addition, this study helped to identify characteristics of patients with poor self-management status. These characteristics can be used to proactively identify patients who might be at risk for poor self-management, including patients with disruptive or socially inappropriate behaviors, impaired decision-making, or smoking. HHC nurses can prioritize such patients for self-management problem assessments and interventions, when necessary.

Future research directions

Several studies reported that symptomatic HF patients who practice effective self-management had a lower risk of events such as mortality, hospitalization, or emergency-room admission during follow-up.^5,62,63 For the next step of this study, we will investigate the association between poor self-management and poor outcomes (e.g., emergency room visits and unplanned readmission).

Limitations

This study has several limitations. First, it should be noted that some clinicians are inconsistent or fail to document poor self-management, and that there may be ethical dilemmas involved in documenting poor self-management, leading us to under-or over-estimate poor self-management. Second, we should bear in mind that the narrative notes may not capture all relevant self-management behaviors. Third, our NLP algorithm may miss some information about patient self-management or may misclassify notes. Second, the EHR data was extracted from a single HHC agency in the Northeastern US, which may limit the generalizability of our findings to other HHC settings.

Conclusions

Patients’ self-management data is often not accessible in a manner that allows clinicians to efficiently guide clinical interventions. This study attempted to overcome this challenge through the use of NLP applied to clinical notes in HF patients. This study identified several patient characteristics associated with poor HF self-management, including male gender, diagnosis of diabetes or depression, impaired decision-making, smoking, and shortness of breath with exertion. These characteristics can help HHC clinicians to identify patients who might be at risk for poor self-management.

Abbreviations:

HF

heart failure

US

United States

HHC

home health care

OASIS

outcome and assessment information set

EHR

electronic health record

EOC

episode of care

AIC

akaike information criterion

OR

odds ratio

CI

confidence interval

ADL/IADL

activities of daily living/instrumental activities of daily living

NLP

natural language processing