Predicting functional adverse outcomes in hospitalized older patients: A systematic review of screening tools

Abstract

Background

Functional decline frequently occurs following hospitalisation in older people and may be prevented or minimized by specific management. Such care processes needs appropriate early screening of older hospitalized patients.

Objective

To identify instruments able to detect on admission older hospitalized patients at risk of functional decline at and after discharge.

Methods

Functional decline is defined as loss of independence in activities of daily living (functional decline) or admission in nursing home. The systematic search used Medline 1970–2007, Web of Science 1981–2007 and references list of relevant papers. An independent epidemiologist assessed methodological quality of the retained articles. independent epidemiologist assessed methodological quality of the retained articles.

Results

We found 12 studies developing predictive tools, including 7145 patients. Functional outcomes were assessed at or after discharge. Preadmission functional status, cognition, and social support were major components for prediction of functional evolution. Few instruments are fully validated and data concerning reliability are often lacking. Operational characteristics are moderate (sensitivity 29–87%, negative likelihood ratio 0.2–0.8).

Conclusions

Instruments predicting functional adverse outcomes are difficult to compare due to heterogeneity of functional outcomes and hospital settings. The reason why so many tools have been developed is probably because none gives full satisfaction: their general predictive validity and performances are insufficient. Further research is needed to improve the screening of frail older patients admitted to hospital with standardized and validated tools.

Introduction

When older people are admitted to hospital for an acute health event, they are at high risk of functional adverse outcomes during and after hospitalisation (1), due to illness severity and pre-existing frailty. However, there is no validated definition of frailty at admission to hospital, and common criteria adopted in community-dwelling patients give heterogeneous results in acute settings (2). Assessment of risk of functional decline may then be proposed as an estimation of frailty since loss of independence is a main outcome of frailty. Increased functional dependence has major impact on quality of life, care of cost and vital prognosis, and appropriate management by the use of comprehensive geriatric assessment improve functional outcomes (3). To plan early intervention, selection of patients is crucial; frail patients are indeed most susceptible to benefit from the geriatric care process (4). Functional decline may be defined directly by increased dependence in Activities of Daily Living (ADL), or indirectly by admission to nursing home. Based on clinical or empirical constructions using risk factors of FD, several instruments predicting loss of function are developed and are more consistent than single variables in their ability to predict adverse outcomes (5). Two previous studies reviewed this subject (5, 6) but they did not focus on predictive instruments. Moreover, they did not include potentially useful tools, either that have been published after the systematic review (5) or were not selected by the search (6). In this systematic review, we aimed to identify previous and newer instruments predicting functional decline in older hospitalized patients and to provide an in-depth analysis of their performances and differences.

Methods

Search strategy

We searched Medline (1970-2007) and Web of Science (1981-2007). The following keywords and MeSH terms were used to identify papers of interest [see appendix]: “aged or aged, 80 and over”, “frail elderly”, “older people”, “hospital”, hospitalisation”, “emergency service”, “predictive value of test”, “prognosis”, “risk assessment”, “geriatric assessment”; “screening”, “predictors”, “functional decline”, “activities of daily living”, “institutionalisation, “patient discharge”.

Additional publications were identified by a manual search of references of relevant papers.

Articles selection

Selection criteria are presented in table 1. All studies included had to be prospective, involve older people (65 years old and over), admitted to hospital. As early detection is a key factor of success in geriatric management, we limited the search to instruments that may be used within the first days of hospitalisation. Patients had to be evaluated with a classification in risk categories with a predictive instrument, and followed at and/or after discharge for functional decline. Functional decline had to be assessed directly (ADL) or indirectly by admission to nursing home. As we were interested in tools applicable in general acute setting, studies restricted to a particular setting were excluded (i.e. heart failure, hip fracture, rehabilitation). One research fellow screened all titles and abstracts identified by the search and selected relevant articles, which were reviewed independently by three investigators with standardised checklist of assessment criteria for choosing reviewed papers.

Table 1.

Selection criteria

Inclusion	Exclusion
65 years and over	Particular setting (e.g. heart failure, hip fracture)
Hospital stay (elective or emergency admission)	Community-dwelling or rehabilitation setting
Cohort study	Determination of predictors / risk factors only
Risk assessment
Early evaluation
Follow-up at and /or after discharge
Functional decline (ADL or NH admission)

Data extraction

The research fellow and the senior author independently abstracted the following data: setting, participants’ characteristics, inclusion and exclusion criteria, evaluation timing, outcomes, follow-up and risk stratification. Performances (e.g. sensibility, specificity, accuracy, likelihood ratio) of the scales were reported from the original paper or estimated on the basis of the data available in the article. Two authors assessed methodological quality of each selected paper with the Oxford Centre for Evidence-based medicine level of evidence rating scale for prognosis studies (7). Score was attributed according to design quality and validation of the clinical diagnosis rule (CDR). Possible scores were: 1a (CDR validated in different populations); 1b (CDR validated in a single population); 1c (all or none case-series); 2a (retrospective cohort studies); 2b (derivation of CDR or validated on split-samples only); 2c (outcomes research); 3: non applicable for prognosis studies; 4 (case -series or low quality cohort studies).

Results

Search results and methodological assessment

Fifty-three relevant articles were fully read by the first author and thirty-one did not meet inclusion criteria. Twenty-two papers were reviewed and discussed by three lectors. Twelve studies describing instruments were retained and their main characteristics are presented in table 2. Only one tool has been validated in different cohorts and reached the highest level of evidence (8). Two studies used separate cohorts for development and validation (9, 10) (level 1b). Others used bootstrapping (11, 12) or removing each individual from the data and re-estimating the parameters of the model (13) to estimate index’s predictive performance. Patients with missing data were excluded from analyses, except in Wu et al. (12) who used imputation strategy. Sampling sometimes includes patients who already had the outcome (nursing home resident) ( 14., 15., 16., 17.). Two tools were tested for interrater reliability (14, 17) and test-retest reliability was reported for a first version of ISAR (27-items) (18).

Table 2.

Setting, participants and level of evidence of the studies

Author*	Level of evidence	Participants	Hospital admission	Acronym	N†	Exclusion	Outcomes
McCusker (8)	1a	65 y +	Non elective	ISAR	D=997 V=676	Too ill, disoriented without proxy, nursing home patients, language	ADL decline or nursing home admission or death
Inouye (9)	1b	70 y+	Medical		D=188 V=142	Severe or terminal illness, disoriented without proxy, LOS < 48h, complete dependence, non medical wards	ADL decline Nursing home admission or death
Sager (10)	1b	70 y+	Non elective	HARP	D=448 V=379	Too ill, admitted for surgery, nursing home patients, complete dependence, death < 3 months	ADL decline Nursing home admission
Cornette (11)	2b	70y+	Non elective	SHERPA	550	Terminal illness, ICU, stroke, LOS < 48h, complete dependence	ADL decline
Wu (12)	2b	80 y+	Non elective		N2m =804 N12m=450	Too ill, admitted for psychiatric or elective surgery, LOS <72h, death <2 months, language	ADL decline
Mateev (19)	2b	65 y+	Medical		198	Terminal illness, stroke, dementia or delirium, nursing home patients, language	Nursing home admission or hospital readmission or rehabilitation or death
Winograd ‡ (4)	2b	65 y+	All		401	Administrative: short LOS, nursing home patients or in geriatric services, geography	Nursing home admission
Zureik (13)	2b	75 y	Non elective		354	Death during hospitalisation, nursing home patients, ICU patients, transfer, non medical ward	Nursing home admission
Narain ‡ (16)	2c	70 y+	Medical	-	Nd=380 N6m=366	-	Nursing home admission
Satish ‡ (15)	2c	65 y+	Medical or surgical	-	507	Geographic, comatose or disoriented without proxy, LOS < 24h	Nursing home admission
Mistiaen (14)	4	65y+	All	BRASS	503	LOS < 48h	Nursing home admission or death
Vandewoude (17)	4	70 y	All	VIP	618	Geriatric ward	Nursing home admission

D=development cohort, V=validation cohort; Nd = at discharge, Nxm= at x months (x=1, 2, 3, 6 or 12 months);

^*Studies are presented according first to their level of evidence and then to the definition of functional decline used (ADL decline or nursing home admission);

^†Number are those available for analysis of outcome of interest;

^‡Studies in veterans care setting: no female or <5%; SHERPA: Score Hospitalier d’Evaluation du Risque de la Perte d’Autonomie; ISAR: Identification of Seniors at risk; BRASS: Blaylock Risk Assessment Screening Score; HARP: Hospital Admission Risk Profile; VIP: Voorloping Indicator voor Plaatsing.

Description of instruments

The common objective of all instruments was to identify older hospitalized patients at risk of adverse health outcomes, and some propose stratification of risk ( 8., 9., 10., 11.). Several studies assessed the patient’s profile for the purpose of discharge planning (9, 13, 14, 17, 19) while others evaluated later outcomes and health care needs (1 to 12 months after discharge) (4, 8, 10., 11., 12., 15, 16, 19). Inclusion criteria concerned all admissions in medical or general wards (4, 9, 14., 15., 16., 17., 19) or were limited to non-elective admissions (8, 10., 11., 12., 13.).

Three different approaches have been used to construct screening tools. Some first identified predictors of FD, and afterwards developed predictive models, algorithms and scoring systems (10-13, 16). Others used literature review, discussion with expert panel (8) in combination with clinical expertise (9, 17, 20) to construct a preliminary tool which was then adapted and tested (8, 17). The third approach counted criteria in a list of geriatric syndromes or functional impairments (4, 15, 19) to determine the best cut-off.

When ADL decline was the outcome of interest, baseline value was the premorbid function defined either as the period before the actual illness (8) or a period of two weeks before admission ( 9., 10., 11., 12.). In studies focussing on nursing home admission, baseline value was the functional status assessed on admission (4, 13., 14., 15., 16., 17., 19).

Instruments were completed with patients or surrogates in most of the cases. One instrument is a self-reporting tool that may be filled in by the patient alone (8). Several tools’ criteria were abstracted from medical records (4, 15). Mean age ranged from 71.4 ± 5.5 (15) to 84.3 ± 5.5 years (13). Main exclusion criteria were administrative (length of stay, language, geographic) in all studies but three (14, 15, 17); impossibility of further decline (nursing home residents (4, 8, 10, 13, 19) or complete dependence (9, 11); severe or terminal illness (8-13, 19); disorientation without available proxy (8, 9, 11, 15, 19).

Functional decline was determined using different ADL scales and cut-offs or considered positive in case of nursing home admission in discharge planning studies. Some authors defined composite outcomes, integrating functional decline (8) or nursing home admission (4, 9, 14, 19) and others outcomes (mortality, readmission, emergency department visits) (table 2).

Predictors

The items of instruments are presented in table 3. The main domains assessed are functional and cognitive (all instruments), social (10/12 instruments), mobility, comorbidity and psychological (6/12 instruments), age and sensorial function (5/12 instruments). One screening instrument introduces a “dynamic” item by asking if patient needs increased assistance since the current illness (8). Only 3 instruments integrated nutritional items (4, 15, 19) and one model used a biological factor (12). When we considered separately ADL decline and nursing home admission, both functional and cognitive predictors were preeminent. Institutionalization was heavily influenced by social factors such principal carer’s wish about patient returning home (13). Two different instruments that were developed separately yielded three identical predictors: age, IADL and cognitive function (10, 11).

Table 3.

Predictors and outcomes

Domains	Predictors	ADL decline		NH admission 9 studies*(ref)	Ratio
		5 studies* (ref)	Ratio
Sociodemographic	Age	(10-12)	3/5	(10, 12-14)	4/9
Biomedical	Active medical problems	(12)	1/5	(4, 13, 14, 16)	4/9
	Previous admissions	(8)	1/5	(14)	1/9
	Polymedication	(8)	1/5	(4, 14, 15)	3/9
	Pain, ulcer	(9)	1/5	(9, 19)	2/9
	Nutrition	-	-	(4, 15, 19)	3/9
	Total	(8, 9, 12)	3/5	(4, 13-16, 19)	6/9
Physical function	ADL	(8, 9, 12)	3/5	(4, 9, 13-17, 19)	8/9
	IADL	(10,11)	2/5	(10, 14, 16, 17, 19)	5/9
	Mobility	(12)	1/5	(4, 14, 15, 19)	4/9
	Falls	(11)	1/5	(4, 15)	2/9
	Sensorial functions	(8)	1/5	(4, 14, 15, 19)	4/9
	Any ADL or IADL impairments	(8-12)	5/5	(4, 9, 13-17, 19)	9/9
Cognition	Summarized questions - delirium	(8, 12)	2/5	(4, 13-16, 19)	6/9
	MMSE	(9-11)	3/5	(9, 10)	2/9
	Any cognitive impairment	(8-12)	5/5	(4, 9, 10, 13-16, 19)	8/9
Psychosocial	Self rated health, QOL	(11,12)	2/5	(12)	1/9
	Depression	(12)	1/5	(4, 12, 15, 19)	4/9
	Behavior pattern	-	-	(14)	1/9
	Low social support and home environment	(9, 12)	2/5	(9, 19)	2/9
	Living location	-	-	(13-17)	5/9
	Socioeconomic problems	-	-	(4, 15)	2/9
	Carer’s wish about patient returning home	-	-	(13)	1/9
	Any social problem	(9, 12)	2/5	(4, 9, 13-17, 19)	8/9
Biological factors	Albumin	(12)	1/5	-	-

^*Two instruments were tested both for ADL decline and nursing home admission (9, 10).

Predictive validity

Some instruments propose several cuts-offs ( 8., 9., 10., 11.) (table 4), which enable adjustment to available resources. Sensitivity and negative predictive value perform moderately and depend on methodological quality. Three instruments achieve the 2b-level of evidence or more, but with very low sensitivity ( 8., 9., 10.). Others positive likelihood ration were similar. Areas under ROC curves were used in 4 studies and ranges between 0.65 and 0.81 (moderate discrimination).

Table 4.

Outcomes, follow-up, score and performances

Author*	Follow-up	Event Rate %	Score range	Cut-off value	Sensitivity (%)	Specificity (%)	+LH	-LH	AUC (95% CI)	Interrater reliability (k)
ADL decline or nursing home admission
McCusker¶ (8)	6 months	29	0-6	≥ 2	74	45	1.4	0.5†	0.66 (0.61-0.71)	0.78**
				≥ 3	48	69	1.6	0.8†
ADL decline
Inouye¶† (9)	Discharge	25	0-4	≥1	88†	54†	1.9†	0.3†	-	-
				≥3	29†	98†	14.5†	0.7†
Sager¶‡ (10)	Discharge	31	0-5	≥2	75†	67†	2.3†	0.4†	-	-
				≥4	32†	89†	2.9†	0.8†
	3 months	18		≥2	80†	46†	1.5†	0.4†	0.65
				≥4	36†	87†	2.8†	0.7†
Cornette (11)	3 months	32	0-11.5	<3.5	85	45	1.5†	0.3†	0.73	-
				<5	68	71	2.3†	0.5†
Wu (12)	2 months	-	Model	-	-	-	-	-	0.81	-
	12 months								0.79
Nursing home admission
Inouye¶‡ (9)		25		≥1	83†	53†	1.8	0.3	-	-
				≥3	22†	96†	5.5	0.8
Sager¶‡ (10)	3 months	3	0-5	≥2	85†	42†	1.5†	0.4†	-	-
				≥4	38†	83†	2.2†	0.8†
Mateev (19)	Discharge	42	0-9	≥2	66†	65†	1.9†	0.5†	-	0.94
	3 months	31			64†	59†	1.6†	0.6†
Winograd (4)	12 months	13	(3 levels)	>1	87†	76†	3.6†	0.2†	-	-
Zureik (13)	Discharge	38	0-9	≥4	74	64	2.1†	0.4†	-	-
Narain (16)	6 months	9	Algorithm	-	84	83	4.9†	0.2†	-	-
Satish (15)	12 months	4	0-12	≥ 1	65†	59†	1.6†	0.6†	-	-
Mistiaen (14)	Discharge	21t	0-40	>10	73†	81†	3.8†	0.3†	-	0.78
Vandewoude (17)	Discharge	2.6	0-3	≥1	81	86	5.8†	0.2†	0.87 (0.82-0.93)	-

⁺LH and-LH: positive and negative likelihood ratios; AUC Area under Receiver Operating Characteristics (ROC) curve (95% Confidence Interval).

^*Studies are presented according first to their level of evidence and then to the definition of functional decline used (ADL decline or nursing home admission).

^¶If study includes two separate cohorts, only the results for the validation cohort are shown.

^†Calculated from data available in original article.

^‡two instruments are presented twice as they presented separate analysis for ADL decline and nursing home admission.

^**Test-restest reliability for a more detailed 27-items screening (18)

Few instruments were validated after development. ISAR score (Identification of Seniors at Risk) has been used in separate dataset to detect severe functional impairment and depression at admission (AUC respectively =0.86, 95%CI=0.75-0.92; and 0.78, 95%CI=0.70-0.84), frequent emergency department visits and frequent hospitalizations (21, 22). ISAR was also used in a randomized control trial for initial screening of eligible patients (23), and compared with others screening tools for patients discharged home from the emergency department (24). In older patients with pneumonia, HARP score (Hospital Admission Risk Profile) was a predictor of 18-months mortality but not of early functional decline (15 days after discharge) (25).

Discussion

In older hospitalized patients, it is important to appropriately address care needs, diagnosis procedures and therapeutic goals early after admission. Identifying patients at-risk of functional decline during and after hospitalization may help us. However, available instruments are highly heterogeneous and they lack of validity, reliability and standardization. Each tool has been designed for specific aim, setting and participants and according to locally available resources.

Studies’ objectives are discharge planning or long-term prediction of risk of adverse functional outcomes. Discharge planning aim to decrease length of stay and improve coordination of services following discharge from hospital (26), and is particularly relevant for hospital management. From an individual point of view, evaluation of long-term risk of functional decline seems more pertinent to identify patients who will really benefit from comprehensive geriatric assessment and preventive interventions.

Setting and participants are also debated. Should screening be limited to non-elective admissions? Patients admitted through the emergency department represent more than half of hospitalized patients and have a higher risk of complications (11); priority should then be given to this subgroup of patients. Age-threshold varies across studies and is expected to change with increasing live expectancy; we suggest that screening should target patients who are 75 years and over. Frailest patients are often excluded from these studies (e.g. nursing home patients, disoriented patients without proxy). Secondary analysis of one study shows that these patients are nevertheless categorized at a high risk of functional decline (27); these patients should deserve a special attention in specific programs.

Major predictors of functional decline are premorbid functional status, cognitive function and specifically, when main outcome is nursing home admission, social factors. It is not surprising that predictors are partially different, as a social outcome may be more heavily influenced by social factors such as carer’s stress (5).

Predictive ability and operational characteristics of these instruments is moderate. To be clinically relevant for screening, sensitivity should at least be higher than 80% and negative likelihood ratio below to 0.5. Only one instrument has been tested in different populations (8) and there may be a publication bias if further validation studies yielded unsatisfactory results and remained unpublished. Data concerning reliability are also lacking.

Both the general approach of these instruments and their individual measures have their limitations. First, all tools are developed using data from a group of patients and are more accurate in predicting outcomes for a group, rather than for the individual patient. They identify low-risk and high-risk patients and may fail for the intermediate-risk group. This limits their use in a clinical setting, where emphasis in on individual care. Secondly, definition of functional decline varies across studies (various ADL scales and thresholds). The appraisal of several domains is required to predict functional status decline, and the aggregation in a score with a rating system is helpful but needs a complex construct. Such assessment is also limited by subjectivity due to human reactions and variations (28). Furthermore we included nursing home admission in our definition of functional decline; decisions on institutional care are indeed rarely made on a purely functional basis, but on mixed functional and social (29). Finally, use of admission data only may limit the predictive discriminatory ability. We had did not include studies collecting data during the hospital stay or at discharge because early interventions require early screening.

McCusker et al. systematically reviewed predictors and predictive indexes of functional decline following hospitalisation and concluded that indexes perform better than single variable. Since this review, other works have been published. Compared with Hoogerduijn et al, we included instruments considering not only decline in function but also nursing home admission as outcome. Furthermore, in that recent work, several instruments predicting functional decline were not included (9, 12, 30). One of the three instruments selected in their work was not specifically designed for older patients and considered “complex care needs” as an outcome, which in our opinion not only reflects functional decline but also illness severity. Finally, we provided an assessment of the methodological quality of each instrument, using Oxford Centre for Evidence-based medicine level of evidence rating scale for prognosis studies (7), and emphasized the relative importance of predictors according to the outcome of interest (functional decline or nursing home admission).

Despite the limitations of instruments predicting functional decline, we think that they may be useful to identify frail older hospitalized patients. Frailty is a state of increased vulnerability resulting from multisystemic reduction of physiological reserves and exposing to deleterious outcomes (functional decline, disability, institutionalization, mortality) (4, 31, 32). If definitions and formal criteria are available for community-dwelling seniors, there is no evidence that these criteria may be useful in hospital setting (2, 33, 34). For this reason, patients who failed to maintain functional homeostasis during and following hospitalization may be considered as frail (35).

How may we improve the screening of elderly hospitalized patients? Standardized assessment in acute care would help recognition and visibility of geriatric medicine, and improve the effectiveness of outcome assessment, quality monitoring and intervention programs (36). In the absence of gold standard, a first step should be to make a choice based on minimum requirements (28): 1) What is the purpose and of the tool? Is the main outcome functional decline, or nursing home admission? 2) Who will use it? A two-step procedure seems appropriate. First, general nursing staff completes a short and easy screening for each older patient. Then, if positive, a specialized geriatric professional performs a more detailed evaluation. 3) Is the tool clinically useful? Instrument should have good face validity. It should be acceptable by the patients and proxies. 4) Is the tool construct adequate? Are items selected from logistic regression or literature, expert panel and clinical experience? Scoring and weighting should reflect the proportional importance of items. 5) Is the tool standardized? The instructions have to be clear, concise and summarized in guidelines for rating. Then, reliability and validity need to be tested in different populations, with comparison of tools’ performances. The ISAR, which is validated in different settings and fulfilling criteria for a short screening tool, is recommended in the current status of knowledge. Finally, physiologic and biologic parameters may improve clinical scoring. Grip strength had been used as marker of frailty in community-dwelling subjects (32) and may also be used in hospital setting (37). Among physical parameters of frailty, gait speed was the strongest predictor of mortality in hospitalized patients with coronary artery disease (33). Yet these physical parameters are not always feasible (i.e. severe illness, disorientation). Potential biological markers include Interleukin-6, C-reactive protein, Tumor necrosis factor- (38). Such biomarkers have been used in community-dwelling cohorts and increased levels were associated with further functional decline and disability (39).

In conclusion, we reviewed instruments predicting functional decline in hospitalized older patients because this is a major hospital outcome and its risk should be appreciated early after admission. Despite the heterogeneity of these instruments in term of objectives, setting and participants, and their lack of validation, they share common criteria (premorbid functional status, cognitive function and social factors). As functional decline is a major consequence of frailty, we propose that such instruments should be operationalized to determine patients’ level of frailty in the context of an acute health event.

Authors’ contributions

All authors participated in the writing and revision of the manuscript. - Design: MSH, CS; - Articles selection: MSH, DS, CS; - Data extraction: MSH, CS; - Methodological assessment: MSH, WD, BB; - Redaction: MSH, PC, CS, BB. All authors have seen and approved the final version of the manuscript.

Conflict of interest

MDSH is supported by a grant from the Walloon Region. Sponsor plays no role in the design, data extraction and redaction of the present article. We declared no conflict of interest. PC, WD, CS are authors of the SHERPA score.

Supplementary data

Systematic search strategy. For MEDLINE: 1. Aged; 2. Frail elderly; 3. Older; 4. Elder*; 5. OR / #1-4; 6. Hospital; 7. Hospitalization; 8. Emergency service; 9. OR/ #6-8; 10. Screening; 11. Prognosis; 12. Predict*; 13. Risk assessment; 14. Geriatric assessment; 15. Predictive value of tests; 16. OR / #10-15; 17. Functional decline; 18. Activities of daily living; 19. Institutionali*; 20. Patient discharge; 21. OR/#17-20; 22. AND /# 5-9-16-21; 23. Limited to: aged and aged, 80 and over. For Web of Science: (Aged or older or elderly) and (hospital* or emergency) and (screening or predict* or risk assessment or outcomes) and (functional decline or activities of daily living or functional impairment)

Financial disclosure

None of the authors had any financial interest or support for this paper.