Depressive Symptoms after Hospitalization in Older Adults: Function and Mortality Outcomes

Abstract

Objectives

To determine the relationship between depressive symptoms after hospitalization and survival and functional outcomes.

Design

Secondary analysis of a prospective cohort study

Setting

General medical service of two urban, teaching hospitals in Ohio

Participants

Hospitalized patients, age 70 years or older.

Measurements

We measured 10 depressive symptoms, instrumental activities of daily living (IADL), and basic activities of daily living (ADL) at hospital discharge and 1, 3, 6, and 12 months later. Using all data points, we determined subject-specific changes in depressive symptoms (slopes). We also defined four groups according to the number of depressive symptoms (≤3 symptoms, Low; 4–10 symptoms, High) at discharge and follow-up: Low-Low, Low-High, High-Low, and High-High. We measured mortality at 3, 6, and 12 months after hospital discharge.

Results

Both subject-specific discharge depressive symptoms and change in depressive symptoms over time (slopes) were associated (P<0.05) with functional and mortality outcomes. At 1 year, more patients in the Low-Low depressive symptom group were alive and independent in instrumental activities of daily living (IADL) and basic activities of daily living (ADL), compared to the Low-High group (49% vs 37%, P=0.02).and more patients in the High-Low group were alive and independent in IADL and ADL, compared to the High-High group (39% vs 19%, P<0.001).

Conclusion

Both the number of depressive symptoms and the change in the number of depressive symptoms during the year after discharge were associated with functional and mortality outcomes in hospitalized older adults. Fewer patients with persistently high or increasing depressive symptoms after hospitalization were alive and functionally independent 1 year later than patients with decreasing or persistently low symptoms, respectively.

INTRODUCTION

Depressive symptoms and disorders are common in hospitalized older adults and are associated with significant morbidity and mortality after discharge. Approximately 10% of older adults hospitalized with medical illness have major depression disorder¹ and up to one-third have significant depressive symptoms.^2,3 Single measures of depressive symptoms in hospitalized older adults are associated with worse quality of life^4,5, function⁶, caregiver mental health⁷, and may be associated with increased mortality.^3,8–11 Older hospitalized adults with depression also use more healthcare resources.^2,12–14 Despite this, little is known about the course of depressive symptoms after hospitalization or their effects on health outcomes.

Recent studies of community-dwelling older adults have identified distinct depressive symptom trajectories^15,16 and persistently high numbers of depressive symptoms were associated with increased mortality.¹⁵ Only one previous study has evaluated depressive symptom trajectories in recently hospitalized patients,¹⁷ but this study did not assess the association of depressive symptom trajectories with patient outcomes.

To address this gap, we measured depressive symptoms at hospital discharge and at 1, 3, 6 and 12 months post-discharge. We first used a repeated measures longitudinal analysis to assess the association between the number of discharge depressive symptoms and the change in the number of depressive symptoms over time with functional and mortality outcomes. Based on the associations demonstrated in the repeated measures analysis, we defined four groups according to the number of depressive symptoms (≤3 symptoms, Low; 4–10 symptoms, High) at discharge and follow-up: Low-Low, Low-High, High-Low, and High-High. In reporting results, we take the perspective of a clinician caring for a patient on discharge from the hospital and ask: Which patient characteristics are associated with worsening or persistently high depressive symptom groups?, and, Do survival and functional outcomes differ (a) between the Low-Low and Low-High depressive symptom groups, and (b) between the High-High and High-Low depressive symptom groups?

METHODS

Patient sample

This study is a secondary data analysis of patients enrolled in prospective studies comparing an intervention aimed at improving functional outcomes to usual care. We studied a random sample of patients aged 70 years or older admitted consecutively to the general medical wards of two study hospitals (University Hospitals of Cleveland and Akron City Hospital) from 1993 through 1997.^18,19 Of 11,475 eligible patients, 3,163 were randomly selected on admission for enrollment and 2,877 agreed to participate. Informed consent was obtained orally from patients or their proxies after randomization and according to procedures approved by the hospital's institutional review board. We combined patients from the intervention and usual care groups as these two groups had similar numbers of depressive symptoms on admission and discharge.

We excluded patients who died during hospitalization (65) or within the first 30 days after hospital discharge (182). Patients who did not complete at least two depression scales (1501) were not included, leaving an analytic cohort of 1129 patients. Patients that were not included were older, more often male, had more comorbidity, higher severity of illness as indicated by the acute physiology component of the Acute Physiology and Chronic Health Evaluation (APACHE) II score,²⁰ and more functional dependency compared to patients included in the analysis.

Data collection

Trained nurse abstractors obtained information from the medical chart. Trained interviewers obtained information from patients at 6 time points: on admission, at discharge, and at 1, 3, 6, and 12 months after hospital discharge. Research nurses were not involved in patient care and were not aware of the research hypothesis.

Primary measurements

Depressive symptoms were measured using the 10-item Center for Epidemiologic Studies Depression (CES-D) scale. Patients were asked if they felt depressed, felt everything was an effort, had restless sleep, felt happy, felt lonely, felt people were unfriendly, enjoyed life, felt sad, felt people disliked them, or could not get going, for much of the time during past week. The range of the scale is 0–10, with one point for each endorsed symptom (positive items were reverse coded). This instrument has been used to evaluate depression in community-living²¹ and hospitalized older adults.^22,23 We used a cutoff of 4 or more depressive symptoms to identify patients with a clinically meaningful number of depressive symptoms since this cutoff demonstrated high sensitivity and specificity in diagnosing major depression in older adults.²¹

Covariates

Sociodemographic characteristics included age, sex, marital status, race, highest educational level, and current living situation. Clinical characteristics included number of admission depression symptoms, comorbid illnesses²⁶ and severity of illness as indicated by the acute physiology component of the APACHE II score. Albumin was measured in g/dl units and was included based on previous work demonstrating its association with failure to recover ADL function in the year after hospitalization²⁷ as well as one-year mortality in hospitalized older adults.¹⁹ Cognitive impairment was measured using the Short Portable Mental Status Questionnaire.²⁸ Patients with >2 errors were classified as having impaired cognitive function. No patient had more than 5 errors in this sample. We measured dependence in bathing, dressing, using the toilet, transferring from bed to chair, and eating using a modified Katz Activities of Daily Living (ADLs) index.²⁹ We measured dependence in 7 instrumental activities of daily living (IADLs): using the telephone, using public transportation or driving, shopping for groceries, preparing, serving and providing meals, doing light housework, taking medications, and managing finances.³⁰ Self-rated global health was measured, with ratings of poor, fair, good, very good, and excellent.³¹

Primary Outcomes

Functional disability was defined as dependency in one or more IADL or ADL at the last report or at one year. We defined the one-year and ten-year mortality outcome as death occurring between one month and one and ten years after hospital discharge, respectively. Vital status at one year and date of death was determined during telephone interviews and verified using the National Death Index.³² Functional and one-year mortality outcomes were combined into four hierarchical, all-inclusive and mutually-exclusive levels: independent in ADL and IADL, independent in ADL and dependent in one or more IADL, dependent in ADL, and dead. Vital status at ten years was determined using the National Death Index.

Missing Data

For most variables, data were missing for fewer than 4% of subjects. There were 4282 opportunities to collect CES-D reports from discharge to the last measurement. Of these assessments, all 10 CES-D items were complete for 67% of assessments, 7–9 items were complete for 14%, 1–6 items were complete for 1%, and no items were completed for 18%. Missing items and assessments were multiply imputed using the ice library.³³ Multiple estimates from the imputed datasets were combined using standard multiple imputation techniques as implemented in the mim library.³⁴ The ice and mim libraries were available through Stata version 10. Sensitivity analyses using subjects with complete assessment of CES-D measures and using the last observation carried forward technique did not change our results. We present results using the imputed CES-D measures.

Analytic strategy

We used a joint modeling approach to assess the relationship between repeated depressive symptom measures and the one-year outcomes.³⁵ We fit a mixed effects linear trajectory model using all CES-D measures to determine subject-specific intercepts (i.e. estimated CES-D score at time of discharge) and subject-specific slopes of CES-D scores over time. Higher order polynomial models for the subject-specific depressive symptom measures over time did not yield substantially better fit. The subject-specific intercepts and slopes were then included as predictors in an ordinal logistic regression using the four-level, one-year outcome variable. This analysis tests the hypothesis that both the number of discharge depressive symptoms and the rate of change in the number of depressive symptoms over time are independently associated with one-year outcomes. The ordinal logistic regression portion of the linear model satisfied the proportional odds assumption (p=0.14). In a separate analysis, we fit an ordinal logistic regression model to assess the association between the number of depressive symptoms at discharge and at last measurement (categorized as low, 0–3 symptoms, and high, 4–10 symptoms) and one-year outcomes.

The joint modeling analyses demonstrated that both subject-specific discharge CES-D measures and slope of CES-D measures over time were independently associated with one-year outcomes. The ordinal logistic regression model demonstrated that both the discharge and last CES-D measures were associated with one-year outcomes. (see Results) To explore these associations further, we classified patients into clinically relevant depressive symptom groups.

Depressive Symptom groups

Patients were classified into four all inclusive, mutually exclusive groups according to the number of depressive symptoms at hospital discharge and at final follow-up; final follow-up was 12 months after discharge for survivors, or at last CES-D measurement before death. Four or more depressive symptoms was defined as a “High” number of symptoms and 3 or fewer symptoms was defined as a “Low” number of symptoms. The four groups were: patients with 3 or fewer depressive symptoms at discharge and follow-up (Low-Low); patients with 3 or fewer depressive symptoms at discharge and 4 or more symptoms at follow-up (Low-High); patients with 4 or more depressive symptoms at discharge and 3 or fewer symptoms at follow-up (High-Low); and patients with 4 or more depressive symptoms at discharge and follow-up (High-High). Unless otherwise noted, analyses and comparisons were made among patients in groups with 0–3 discharge depressive symptoms (Low-Low vs. Low-High) and, separately, among patients in groups with 4–10 depressive symptoms (High-High vs High-Low). Boxplots of CES-D scores over time were plotted. We used χ² tests for categorical variables and the Wilcoxon test statistic for continuous variables to examine differences in patient characteristics between depression symptom groups. The groups did not differ (p>0.1) in the proportions of patients in the intervention and usual care group in the original studies. We calculated logistic regression models to examine the association between patient characteristics and depressive groups. We used partial proportional odds models to examine the association between depressive symptom groups and the one-year outcome as the proportional odds assumption was not met for ADL and IADL dependencies. Partial proportional odds modeling³⁶ relaxes the proportional odds assumption of standard ordinal logistic regression modeling for some of the predictors. The partial proportional odds model included potential mediators such as discharge depressive symptoms, IADL and ADL dependencies at discharge, and a validated one year mortality risk index for hospitalized older adults¹⁹. We completed an additional analysis comparing one-year outcomes amongst all 4 groups and described all pairwise comparisons. Cox proportional hazards models were used to examine the association of subject-specific slopes and discharge depressive symptoms, and separately, depression groupings, with long-term mortality up to 10 years post-admission. The proportional hazards assumption was met for the covariates in these 10-year mortality models. The median survival time was 5.0 years (95% CI 4.4–5.4). Analyses were performed using SAS version 9.2 or Stata version 10.

RESULTS

The study cohort (Table 1) included 1129 patients with a mean age of 78.2 years; 66% were women, 22% African-American, and 44% were married. The most common comorbid conditions were congestive heart failure, chronic lung disease, diabetes mellitus, and peripheral vascular disease. On admission, most patients were independent in ADL and had 3 or fewer depression symptoms. One hundred thirty three patients (12%) died in the year after discharge. Of these, 19% died between 1 and 3 months, 30% died between 3 months and 6 months, and 51% died between 6 months and 12 months. An additional 733 (65%) patients died within 1 and 10 years after discharge.

Table 1.

Patient Characteristics^* (N=1129)

Sociodemographic	n (col %) or Mean (SD)†
Age in years, Mean ± SD	78.2 ± 5.9
Women	744 (66)
Race
Non-White	243 (22)
Marital Status
Married	494 (44)
Widowed	158 (14)
Education, yr
<12 years	415 (38)
Lives alone	459 (41)
Comorbidity/Impairments
Comorbid conditions
Congestive Heart Failure	289 (26)
Chronic Lung Disease	265 (24)
Diabetes	207 (18)
Peripheral Vascular Disease	205 (18)
Cerebrovascular Disease	164 (15)
Past Myocardial Infarction	162 (14)
Metastatic Cancer	27 (2)
Charlson Comorbidity Score, Mean ± SD	1.9 ± 1.8
Serum Albumin in g/dl, Mean ± SD	3.6 ± 0.6
Acute Physiology Score, Mean ± SD	9.7 ± 3.1
Function
Admission
IADL
0–2 dependencies	516 (46)
3+ dependencies	607 (54)
ADL
0 dependencies	608 (54)
1 dependency	176 (16)
2+ dependencies	340 (30)
Cognitive function
Normal	873 (78)
Discharge
IADL
0–2 dependencies	715 (63)
3+ dependencies	414 (37)
ADL
0 dependencies	759 (67)
1 dependency	170 (15)
2+ dependencies	200 (18)
Depression symptoms
Admission
Mild (0–3)	735 (65)
Moderate (4–6)	282 (25)
Severe (7–10)	112 (10)
Discharge
Mild (0–3)	719 (64)
Moderate (4–6)	305 (27)
Severe (7–10)	105 (9)
Self rated health
Poor/Fair	358 (46)
Good/Very Good/ Excellent	601 (53)

^*Data were missing for the following characteristics: education (n=42); living alone (n=13); comorbid conditions (n=6); Charlson comorbidity score (n=6); Acute Physiology Score (n=3); albumin (n=116); admission Instrumental activity of daily living (IADL)(n=6); admission Activity of daily living (ADL)(n=5); admission cognitive function (n=3); self rated health (n=170).

^†some percents do not add to 100% due to rounding

Results of the joint mixed effects and ordinal logistic regression modeling demonstrated that both the number of depressive symptoms at discharge and the change in the number of depressive symptoms over time were independently associated with the one-year, hierarchical outcome of alive and independent, dependent in IADL and independent in ADL, dependent in ADL, or dead. The one-year outcome was associated with the number of depressive symptoms on discharge (OR 1.25; 95% CI, 1.18, 1.33; P<0.001) and with the subject-specific slopes of CESD scores after discharge (for each additional SD increase [equal to 0.5 additional symptom per year], OR, 1.26; 95% CI, 1.09, 1.46; P= 0.002).

Results were confirmatory in an ordinal logistic regression model of the one-year, hierarchical outcome of alive and independent, dependent in IADL and independent in ADL, dependent in ADL, or dead with discharge CES-D score and last CES-D score. The one-year outcome was independently associated with 4–10 depressive symptoms, compared to 0–3 depressive symptoms at discharge (OR, 1.5; 95% C.I., 1.2–1.8; P=0.002), and at last measurement (OR, 2.6; 95% C.I., 1.9–3.4; P<0.001).

In a proportional hazards model of 10-year mortality, the 10-year outcome was associated with the number of depressive symptoms at discharge (HR 1.11; 95% CI [1.07,1.15], P<0.001) and with the subject-specific slopes of CES-D scores from discharge to 1 year later (for each additional SD increase [equal to 0.5 additional symptom per year], HR 1.14; 95% CI [1.05, 1.25]; P= 0.003).

Characteristics of depressive symptom groups

Among the 719 patients with 0–3 depressive symptoms at discharge (Figure 1A), 643 (89%) had 0–3 symptoms one year later (the Low-Low group) and only 76 patients reported 4 –10 depressive symptoms one year later (the Low-High group). Among the 410 patients with 4–10 depressive symptoms at discharge (Figure 1B), 258 (63%) had 0–3 symptoms one year later (the High-Low group) and 152 reported 4–10 depressive symptoms (the High-High group).

Figure 1. Depressive Symptom Group during study follow-up.

The top panel represents patients with 0–3 depression symptoms at discharge (n=719). The bottom panel represents patients with 4–10 depression symptoms at hospital discharge (n=410). Data shown are actual means (dots) and medians (horizontal bars) of Center for Epidemiologic Studies Depression (CESD) Scale symptoms over time. The mean number of depressive symptoms for the Low-Low group are 1.30, 1.14, 1.16, 1.02, and 0.77; for the Low-High group, 1.79, 2.75, 2.86, 2.46, 5.37; for the High-High group, 5.92, 4.84, 4.72, 4.40, and 5.32; and for the High-Low group, 5.32, 2.33, 2.06, 1.55, and 1.36,, at discharge, 1 month, 3 months, 6months, and 1 year, respectively. Boxes represent the interquartile range of the number of depressive symptoms for each group over time. The width of each box is proportional to the number of patients. Vertical dashed lines represent the range of values encompassing 5% to 95% of depressive symptoms for each group over time.

Characteristics of patients in different depressive symptom groups

Among patients with 0–3 depressive symptoms at discharge, those in the Low-High group had more depressive symptoms on admission, and were more likely to live alone, have diabetes, or report fair or poor health, compared to patients in the Low-Low group (Table 2a); these differences remained significant in a multivariate logistic regression model. Among patients with 0–3 depressive symptoms at discharge, those without diabetes who live with an adult child, and report no depressive symptoms on admission and good general health, have a 96% probability of being in the Low-Low group and a 4% probability of being in the Low-High group.

Table 2a.

Characteristics of Patients with 0–3 Depressive Symptoms at Hospital Discharge according to Group and Comparisons between Groups^*

Characteristics, (col % or mean (SD)	Low-Low Group n=643	Low-High Group n=76	Unadjusted bivariate p-values	Adjusted Multivariate results OR (95% CI)†
Sociodemographic
Age in years, Mean ± SD	77.7 ± 58	78.5 ± 6.3	0.38
Women	65	74	0.11
Race			0.41
Non-White	23	18
Marital status			0.19
Married	47	36
Education			0.71
≥ 12 years	62	66
Lives alone	37	51	0.02	1.8 (1.1–3.0)
Comorbidity/Impairments
Comorbid conditions
Diabetes	19	32	0.01	1.8 (1.1–3.1)
Charlson Comorbidity Score, Mean ± SD	1.8 ± 1.8	2.3 ± 2.0	0.05
Serum Albumin in g/dl, Mean ± SD	3.6 ± 0.5	3.5 ± 0.5	0.22
Acute Physiology Score, Mean ± SD	9.4 ± 3.0	10.1 ± 3.3	0.09
Function
Admission
Cognitive function			0.96
Normal	80	80
Discharge
IADL, Mean SD	1.5 ± 1.8	1.8 ± 1.9	0.15
ADL, Mean SD	0.5 ± 1.1	0.7 ± 1.1	0.13
Depression Symptoms
Admission CESD Score	1.9 ± 1.9	2.9 ± 2.5	<0.0001	1.2 (1.1–1.4)
Self rated health
Poor/Fair	26	46	<0.001	2.0 (1.2–3.4)
Intervention Group			0.69
Usual Care	48	50

^*IADL=instrumental activity of daily living, ADL=Activity of Daily Living, CESD = Center for Epidemiologic Studies Depression Scale-10.

^†The variables that were significant at p<0.05 were used in the multivariate logistic regression model. The Low-Low Group is the referent group.

Among patients with 4–10 depressive symptoms at discharge, those in the High-High group had more depressive symptoms on admission; fewer than 12 years education; and were dependent in more IADLs and ADLs at discharge compared to patients in the High-Low group (Table 2b). Among patients with 4–10 depressive symptoms at discharge, a married person with a high school education, three depressive symptoms on admission, and no IADL or ADL dependencies at discharge, has an 85% probability of being in the High-Low group and a 15% probability of being in the High-High group.

Table 2b.

Characteristics of Patients with 4–10 Depressive Symptoms at Hospital Discharge according to Group and Comparisons between Groups^*

Characteristics, n (col %)	High-Low Group n=258	High-High Group n=152	Unadjusted bivariate p-values	Audjusted Multivariate results OR (95% CI)†
Sociodemographic
Age in years, Mean ± SD	78.7 ± 6.0	78.8 ±5.6	0.82
Women	61	76	<0.01	1.2 (0.7–2.0)
Race			0.08
Non-White	18	25
Marital status			0.01	0.6 (0.4–1.1)
Married	42	28
Education			0.04	0.6 (0.4–0.9)
≥ 12 years	62	66
Lives alone	40	51	0.04	1.1 (0.7–1.9)
Comorbidity/Impairments
Comorbid conditions
Diabetes	15	15	0.96
Comorbidity Score, Mean ± SD	1.8 ± 1.7	1.9 ± 1.6	0.45
Serum Albumin in g/dl, Mean ± SD	3.5 ± 0.6	3.6 ± 0.6	0.12
Acute Physiology Score, Mean ± SD	10.0 ± 3.3	10.2 ± 3.0	0.34
Function
Admission
Cognitive function			0.97
Normal	74	74
Discharge
IADL, Mean SD	2.4 ± 2.0	3.3 ± 2.1	<0.0001	1.1 (1.0–1.2)
ADL, Mean SD	0.8 ± 1.4	1.4 ± 1.6	<0.0001	1.2 (1.0–1.4)
Depression Symptoms
Admission CESD Score	3.6 ± 2.4	5.1 ± 2.6	<0.0001	1.3 (1.2–1.4)
Self rated health
Poor/Fair	40	56	<0.01	1.2 (0.7–1.9)
Intervention Group			0.59
Usual Care	51	45

^*IADL=instrumental activity of daily living, ADL=Activity of Daily Living, CESD = Center for Epidemiologic Studies Depression Scale-10.

^†The variables that were significant at p<0.05 were used in the multivariate logistic regression model. High-Low group is the referent group.

In other pairwise comparisons, patients in the Low-High group had higher prevalence of diabetes, fewer IADL dependencies at discharge, and lower admission CES-D scores, than patients in the High-High group, in a multivariate adjusted model. Patients in the Low-Low group had fewer IADL dependencies, and lower admission CES-D scores than those in the High-Low group.

Functional and mortality outcomes

More patients in the Low-High group were dependent in IADL or ADL or dead within one year of discharge (50% vs 33% in the Low-Low group, p=0.007) and this association remained significant (49% vs 37%, respectively, p=0.02) after adjusting for discharge depressive symptoms, discharge IADL and ADL function, and a one year mortality risk index (Table 3). More patients in the High-High group were dependent in IADL or ADL or dead within one year of discharge, (44% vs 16% in the High-Low group, p<0.001) and this association remained significant (39% vs 19%, respectively, p<0.001) after adjusting for discharge depressive symptoms, discharge IADL and ADL function, and a one year mortality risk index. For other pairwise comparisons, the Low-High group had worse outcomes compared to the High-Low group (p=0.02), the Low-High group had worse outcomes compared to the High-High group (p=0.05), the Low-Low group had similar outcomes to the High-Low group (p=0.78), and the Low-Low group had worse outcomes compared to the High-High group (p<0.001).

Table 3.

Functional and 1 Year Mortality Outcomes by Group (N=1129)

	Alive Independe nt	Alive Dependent in IADL Independent in ADL	Alive Dependent in ADL	Dead	p-value*
0–3 Depression Symptoms at Discharge
Bivariate Analyses	Unadjusted Percents
Low-Low Group (n=643)	50%	22%	20%	8%
Low-High Group (n=76)	33	28	26	13	0.007
Multivariate Analyses, adjusting for Discharge characteristics†	Adjusted Predicted Probabilities
Low-Low Group (n=643)	0.49	0.27	0.19	0.06
Low-High Group (n=76)	0.37	0.28	0.26	0.09	0.02
4 –10 Depression Symptoms at Discharge
Bivariate Analyses	Unadjusted Percents
High-Low Group (n=258)	44%	24%	20%	12%
High-High Group (n=152)	16	20	38	25	<0.001
Multivariate Analyses, adjusting for Discharge characteristics†	Adjusted Predicted Probabilities
High-Low Group (n=258)	0.39	0.29	0.24	0.09
High-High Group (n=152)	0.19	0.24	0.36	0.21	<0.0001

^*P-value for association of low-low vs low-high and high-low vs high-high, respectively with ordinal outcome in a partial proportional odds model. P-value for association of all 4 groups with ordinal outcome in an adjusted partial proportional odds model is <0.001.

^†Discharge characteristics included in the partial proportional odds models were depression symptoms, ADL and IADL dependencies, intervention group, and one-year post-hospitalization mortality risk index. This index uses the following point scoring system: male sex-1 point, 1–4 ADL dependencies-2 points, 5 ADL dependencies-5 points; presence of congestive heart failure-2 points, cancer-3 points, metastatic cancer-8points; creatinine >3.0 mg/dl-2 points; albumin between 3.0–3.4 g/dl-1 point and <3.0 g/dl-2 points.

At 10 years, patients in the Low-High group were no more likely than patients in the Low-Low group to have died (77% vs 73%, respectively, p=0.15, hazard ratio (HR) = 1.2, 95% confidence interval (C.I.) =0.9–1.6) after adjusting for age, and discharge depressive symptoms, IADLs, and ADLs. At 10 years, patients in the High-High group had a higher risk of mortality (91% vs 79% in the High-Low group, p<0.0001, HR-1.7, 95% C.I.=1.3–2.1) after adjusting for age and discharge depression symptoms, IADLs, and ADLs.

DISCUSSION

We emphasize three findings from a one-year, longitudinal analysis of depressive symptoms in older patients discharged from the hospital. First, in repeated measures longitudinal analyses, both the number of depressive symptoms and the change in depressive symptoms over time were important clinical predictors of poor outcome. Second, in an exploratory analysis of clinically relevant depressive symptom groups, more than half of all patients were in the Low-Low group and nearly one quarter of all patients were in the High-Low group. Third, the Low-High and High-High groups had signs of preexisting vulnerability, such as more depressive symptoms on admission, worse general health or function, and living alone and had worse functional outcomes and survival, compared to the Low-Low and High-Low groups, respectively.

Only 7% of patients were in the Low-High group. Compared to the Low-Low group, the Low-High group was associated with signs of pre-existing vulnerability and worse one year functional outcomes and survival. This difference was moderate in size, statistically significant, robust across outcomes, and consistent in multivariate analyses, but did not extend to higher ten-year mortality. Our analyses did not determine whether increasing depressive symptoms preceded worsening ADL and IADL function, and it is possible that depressive symptoms are an epiphenomenon of poor or declining health.

Most patients with 4–10 depressive symptoms at discharge were in the High-Low group. The decline in symptoms occurred early, a finding also reported by Koenig et al.³⁷ The High-High group was associated with signs of preexisting vulnerability and had strikingly poor outcomes compared to the High-Low group; with 25% dying and only 16% independent in all ADLs and IADLs 1 year after discharge and higher risk for death at 10 years. Our findings are consistent with the hypothesis that depressive symptoms worsen outcomes. While function and depressive symptoms were measured concurrently, mortality was measured after the last depressive symptom measurement. Thus, the association of depressive symptoms with mortality indicates this is not simply a cross-sectional association. Previous studies have differed as to whether depressive symptoms in hospitalized patients are associated with mortality after discharge¹¹. However, these studies did not determine whether symptoms persisted after discharge. We found worse outcomes in the Low-High group, compared to the Low-Low group, and in the High-High group compared to the High-Low group. Thus, persistent or increasing depressive symptoms are associated with worse outcomes.

The results of the joint modeling analysis and the association of worse functional and mortality outcomes in the High-High compared to the Low-High group support the intriguing possibility that the trajectory of depressive symptoms is important in determining functional and mortality outcomes. This was not seen in the comparison of the Low-Low and High-Low groups and this question should be a topic of further research.

Implications

Our study has several implications. First, most hospitalized older adults, even those with many depressive symptoms at discharge, report few depressive symptoms a year after discharge. Second, assessment for depressive symptoms and other specific signs of vulnerability on admission and discharge can identify patients at increased risk for 4 or more depressive symptoms a year later. Third, patients in groups that reported an increasing or persistently high number of depressive symptoms after hospitalization were independently associated with worse functional outcomes and mortality after hospitalization, even when controlling for prognostic factors such as comorbid illness, severity of illness, and functional status. For those with persistent depressive symptoms, this association extended to increased mortality at 10 years, a finding also seen in hospitalized patients based on a single measure of depressive symptoms.¹⁰ Although it is unclear whether the effects of depressive symptoms on health outcomes are direct or indirect, worsening or persistent depressive symptoms may have adverse physiologic effects,³⁸ reduce physical activity,³⁹ and decrease appropriate medical care. Thus, our findings provide a rationale for measuring depressive symptoms in hospitalized older adults, especially because they are often unrecognized.^40–42 Finally, physicians can use our findings to identify high-risk hospitalized older adults and to target them for intensive management. While previous interventions based on a single measurement of depressive symptoms have been disappointing,^{43, 44} a recent intervention targeting hospitalized patients with persistent depressive symptoms has demonstrated promising results.⁴⁵

Methodological Considerations

Several methodological considerations support the validity of our findings including, the large size of the cohort; the ability to evaluate potential confounders, such as comorbid illness, functional status, and cognitive status; use of a valid measure of depressive symptoms; and near complete ascertainment of outcomes

We recognize important limitations of our study. First, our findings may not apply to patients who were too sick to participate in the measurement of depressive symptoms at discharge or to surgical patients, who were not included in our study. Also, patients were recruited from two large urban hospitals and may not reflect patient populations in other settings. Second, we measured depressive symptoms and did not classify patients according to Diagnostic and Statistical Manual of Mental Disorders-IV (DSM-IV) diagnoses. We do not propose the use of the CES-D as a diagnostic instrument for depression,²³ but, the strong association of depressive symptoms, as measured with the CES-D, and poor outcomes demonstrated in this and other studies^46,47 highlight their importance. Third, antidepressant use after hospitalization was not measured and we are unable to determine the effect of antidepressant use on groups and outcomes. Fourth, hospital readmission and other clinical events occurring six or more months after hospitalization were not recorded. Thus, we cannot determine whether these influenced depressive symptom group. Fifth, our analyses had concurrent measurement of depressive symptoms and functional outcomes, so we could not determine the temporal relationship. However, the association of depressive symptoms with mortality indicates this is not simply a cross-sectional association. In community-dwelling older adults, Barry et al. demonstrated the temporal precedence of depressive symptoms and subsequent disability and failure to recover from disability.⁴⁸ Future work should assess the temporal sequence of these factors and also measure putative mediators such as physical activity and use of health care.

In conclusion, our results show that both the number of discharge depressive symptoms and the change in depressive symptoms after discharge are important in older adults discharged from the hospital. Patients who demonstrate persistently elevated or increasing depression symptoms have poor functional and mortality outcomes. Moreover, patients at high risk for developing persistently elevated or increasing depression symptoms can be identified based on clinical features at admission. These results provide a rationale to monitor depressive symptoms during and after hospitalization and demonstrate a need for interventions that can reduce symptoms and improve function and survival among at-risk patients.