Risk Factors for Early Hospital Readmission in Patients with AIDS and Pneumonia

Abstract

OBJECTIVE

To determine risk factors for early readmission to the hospital in patients with AIDS and pneumonia.

DESIGN

Case-control analysis.

SETTING

A municipal teaching hospital serving an indigent population.

PATIENTS

Case patients were all AIDS patients hospitalized with Pneumocystis carinii pneumonia or bacterial pneumonia between January 1992 and March 1995 who were readmitted for any nonelective reason within 2 weeks of discharge (n = 90). Control patients were randomly selected AIDS patients admitted during the study period who were not early readmissions (n = 87), matched by proportion of Pneumocystis carinii to bacterial pneumonia.

MEASUREMENTS AND MAIN RESULTS

Demographics, social support, health-related behaviors, clinical aspects of the acute hospitalization, and general medical status were the main predictors measured.

RESULTS

Patients were at significantly increased risk of early readmission if they left the hospital unaccompanied by family or friend (odds ratio [OR] 4.76; 95% confidence interval [CI] 2.06, 11.0; p = .0003), used crack cocaine (OR 3.40; 95% CI 1.02, 11.3; p = .046), had one or more coincident AIDS diagnoses (OR 3.65; 95% CI 1.44, 9.26; p = .0065), or had been admitted in the preceding 6 months (OR 2.82; 95% CI 1.21, 6.57; p = .016). Demographic characteristics, alcoholism, intravenous drug use, illness severity on admission, and length of hospitalization did not predict early readmission.

CONCLUSIONS

Absence of companion at discharge and crack use were important risk factors for early readmission in patients with AIDS and pneumonia. Additional AIDS comorbidity and recent antecedent hospitalization were also risk factors; however, demographics and measures of acute illness during index hospitalization did not predict early readmission.

As the cost of health care has escalated, health care systems, insurers, and policy makers have sought to reduce expenditures by reducing days of hospitalization. Because early hospital readmission is both expensive and potentially avoidable, substantial savings could be achieved by preventing early readmissions, either by improving the quality of care during the initial hospitalization or by identifying patients most likely to be readmitted and modifying their care to reduce their readmission risk.^1,2

Although improving the quality of hospital care is an important goal,^3–5 a very small percentage of early readmissions are directly attributable to substandard care.⁶ Moreover, a recent meta-analysis has emphasized the methodologic difficulty of measuring the process of hospital care.⁷ An alternative strategy is to characterize risk factors—particularly clinical risk factors evident during the index admission—that identify a subset of patients at highest risk of early readmission.

Studies have examined risk factors for early readmission in general medicine patients, veterans, intensive care unit patients, the elderly, and patients with cardiac disease.^8–13 Thus far, there have been no published studies focusing exclusively on patients with AIDS. One study of a general medical population found that AIDS itself was a risk factor for readmission,¹⁴ and one abstract reported that AIDS patients with non-Hodgkin's lymphoma, shorter hospital stays, and veterans benefits were more likely to be readmitted.¹⁵ Because patients with AIDS increasingly include marginalized members of society such as drug users and the homeless, risk factors drawn from previous readmission studies (which involve mostly middle-aged white males) may not be applicable.

To determine risk factors for early readmission, we performed a case-control analysis of patients with AIDS admitted for either Pneumocystis carinii pneumonia (PCP) or bacterial pneumonia (BP) who were readmitted within 14 days. We hypothesized that in this patient population social factors (such as social support, access to care, and drug use) would play an important role in predicting risk of early readmission. Our goal was to define a subgroup of patients who might require further or alternative interventions and suggest the nature of such interventions.

METHODS

The study was conducted at the San Francisco General Hospital (SFGH), a 325-bed county hospital affiliated with the University of California, San Francisco. This hospital is the city's primary source of care for its indigent population and its major center for AIDS care.

In this retrospective case-control study, all admissions from January 1, 1992, to March 31, 1995, involving patients with AIDS and either PCP or BP (excluding mycobacterial pneumonias) were retrieved from the central computer database using discharge diagnosis codes. Case patients were defined as any patients with AIDS discharged from the hospital with a diagnosis of PCP or BP who were subsequently readmitted to SFGH within 14 days for any nonelective reason. Control patients were randomly selected in a 1:1 case-control ratio from the remaining patients who were not readmitted within 14 days. We excluded patients in whom the index hospitalization ended in death or discharge to terminal hospice care. Potential control patients were excluded if they had no further contact with SFGH or its outpatient clinics in order to reduce the possibility that a control patient may have actually been admitted elsewhere within the 14-day window period. One patient with Rhodococcus equi was excluded because repeated hospitalizations were required to identify and treat this rare pathogen.

Case and control patients were matched by proportion of PCP and BP index hospitalizations. Most PCP diagnoses (92%) were microbiologically confirmed, with the remaining diagnoses based on clinical history, radiographic findings, and response to therapy. Bacterial pneumonias were diagnosed by clinical presentation, evidence of infiltrate on chest radiography, and appropriate response to antibacterial therapy. One control patient admitted with concurrent PCP and PB was randomly assigned to the BP group. In patients hospitalized more than once during the study period, only one (randomly chosen) hospitalization was included for analysis.

Data pertaining to the index hospitalization were collected directly from the hospital chart using a standardized data recording form, except for Visiting Nurses Association (VNA) case management records, which were obtained from the VNA computer database. Data recorded included demographic, socioeconomic, and health-related behavior information, as well as data pertaining to the patient's index hospitalization and general medical condition. The study was conducted in the era that largely predated highly active antiretroviral therapy and routine measurement of HIV viral loads.

Data collected from the hospital computer database were available for all case and control patients. Of the 114 case patients identified, 90 (79%) were available for chart review and are the subject of this analysis. Similarly, of the 111 randomly selected control patients, 87 (78%) were available for chart review and are reviewed herein. During the period of chart review, the unavailable charts were missing from the central medical records department and most likely were checked out to other physicians or misfiled. There were no significant differences between reviewed and missing charts based on the available administrative data (e.g., age, ICD-9 diagnosis category, and number of hospitalizations). Categorical data obtained from chart review were analyzed by Mantel-Haenszel χ²odds ratios (ORs). Wilcoxon Two-Sample Test (normal approximation) was used for continuous variables. All statistics comparing case and control patients were analyzed both for the entire group and by type of pneumonia (the matching variable). Significant risk factors identified by univariate analysis (p< .05) were further analyzed by multivariate logistic regression. Odds ratios and 95% confidence intervals (CIs) were calculated. All statistical analysis was performed using the SAS computer program (SAS Institute, Cary, NC).

RESULTS

Between January 1, 1992, and March 31, 1995, 852 patients with AIDS accounted for 2,376 admissions to SFGH for PCP or BP. Of the 1,561 PCP cases, 65 (4.2%) had a subsequent readmission within 14 days; and of the 815 BP cases, 53 (6.5%) had a subsequent early readmission. Of the 852 patients admitted during this time period, 13.6% required at least one early readmission.

Case and control patients were matched by proportion of PCP and BP. There were 54 PCP and 36 BP case patients matched to 54 PCP and 33 BP control patients. There were no significant differences between cases and control patients in age, gender, race, CD4 count, length of initial hospitalization, use of antiretroviral medications, case management by VNA, or severity of acute illness (Table 1) All case patients were readmitted within 14 days (median, 7 days; range, 1–14 days). Among control patients, the median time to readmission was 148 days (range, 17–1,044 days). This median value includes 10 censored records, for which the final day of the study was used as the readmission day. There were also nine records for which the readmission date was known only to be greater than the 14-day early readmission window. When those missing records were assigned a value of 15 days, the median time to readmission for all control patients was 134 days.

Table 1.

Characteristics of Case and Control Patients ^*

Characteristics	CasePatients(n= 90)	ControlPatients(n= 87)
Median age, years	38	35
Male, %	82	84
Nonwhite, %	46	36
Unmarried, %	89	86
HIV transmission, same sex, %	50	56
Median CD4 count, cells/mm3	19	26
Median days in hospital	7	6
Prior use of any anti-HIV medicines, %	61	68
VNA case managed, %	32	30

^*There were no significant differences between case and control patients (by Fisher's Exact Test, p< .05).

Social Factors and Health-Related Behaviors

The most predictive risk factors for early readmission were associated with social network and health-related behavior. Patients who left the hospital alone at time of initial discharge were significantly more likely to be readmitted within 14 days than those patients who were accompanied by family or friends when leaving the hospital (Table 2) This information, which was coded as part of the Nursing Discharge Summary, was available for 71 (79%) of 90 case patients and 59 (68%) of 87 control patients. Although roughly one quarter of the Nursing Discharge Summaries did not indicate how the patient left the hospital, discussion with several of the nurses who filled out these forms revealed that this omission was generally due to random oversight. To determine the possible effect of excluding these missing records, we recalculated ORs with the missing records counted either entirely as “leaving alone” or as “leaving with someone.” For both extremes, the finding remained significant (OR 2.69; 95% CI 1.39, 5.22; and OR 3.76; 95% CI 2.02, 6.99, respectively). Of note, mode of leaving the hospital was the only predictor variable for which there were missing records (Table 3).

Table 2.

Predictors of Early Readmission: Univariate Odds Ratios by Pneumonia Type ^*

	PCP	Bacterial Pneumonia	Combined
Left hospital alone	5.54 (2.17, 14.1; .001) †	2.86 (0.86, 9.45; .085)	4.29 (2.07, 8.92; .001) †
Crack use	4.26 (1.20, 15.12; .025) †	2.50 (0.70, 8.91; .16)	3.27 (1.35, 7.93; .009) †
>1 Coincident AIDS diagnosis	1.73 (0.76, 3.95; .192)	6.32 (1.96, 20.4; .002) †	2.69 (1.34, 5.22; .003) †
CD4 < 50	0.94 (0.41, 2.15; .89)	3.83 (1.13, 13.0; .03) †	1.48 (0.76, 2.89; .25)
Previous admission <6 months	1.45 (0.66, 3.22; .352)	4.29 (1.56, 11.7; .004) †	2.16 (1.18, 3.94; .012) †

^*Values are given as odds ratios (95% confidence intervals; p values). PCP indicates Pneumocystis carinii pneumonia.

^†p< .05, Fisher's Exact Test.

Table 3.

Predictor Variables by Number ofCase and Control Patients

	Case Patients		Control Patients
Variable	n	Total	n	Total
Left hospital alone	53	71	24	59
Crack use	20	90	8	87
>1 Coincident AIDSdiagnosis	72	90	52	87
CD4 < 50	69	90	60	87
<6 Months sinceprevious admission	49	90	31	87

Interestingly, leaving the hospital alone was not correlated with living alone. We found no significant differences in living situation between case and control patients, either before or after the index admission. Although not statistically significant, it is noteworthy that 13 (65%) of 20 patients homeless at time of index admission were readmitted within 14 days, including all 5 patients designated as homeless at time of discharge. Other measures of access to care and quality of ambulatory care—type of insurance, VNA case management, identification of an emergency contact or of a primary physician, and scheduling of an outpatient appointment at time of discharge—were not significant. This finding implies that leaving the hospital alone is a more sensitive measure of social isolation than other, more common measures. Employment status did not reach statistical significance, although fewer case patients (6 [6.7%] of 90) than control patients (14 [16%] of 87) were employed at time of initial admission.

Of the health-related behaviors, crack cocaine use was found to be a significant risk factor for early readmission (Table 2), while intravenous drug use and alcoholism were not. Twenty-two percent of case patients but only 8% of control patients were active crack users. There was no significant difference in current or previous extent of cigarette smoking.

When leaving the hospital alone and crack use were analyzed by types of pneumonia, they remained significant in the PCP subgroup. The ORs remained greater than 1.0 in the smaller BP subgroup, but they no longer reached statistical significance (Table 2).

The Initial Hospitalization

There were no significant differences in measured clinical variables at the time of initial hospitalization. Case and control patients had similar Acute Physiology Scores on admission and discharge. (The Acute Physiology Score, often applied to patients in the intensive care unit, is a scale that assigns a severity-of-illness score based on the extent of a patient's abnormalities of temperature, blood pressure, pulse rate, respiratory rate, blood pressure, and oxygen saturation.) Similarly, there were no significant differences in laboratory values (electrolytes, albumin, hematocrit, white blood cell count, and platelets), days of hospitalization, days in the intensive care unit, or days intubated. Use of steroids, requirement for methadone or benzodiazepines for drug withdrawal, admission to the dedicated AIDS unit, and consultation by the AIDS Substance Abuse Program or the AIDS Medical Consult Service likewise were not significantly different between case and control patients. More of those requiring blood transfusion were case patients (21 [65%] of 32, p= .08).

General Medical Status

Patients in both the case and control groups were in the advanced stages of AIDS. Their CD4 counts were similar (mean, 0.034 vs 0.042 × 10⁹/mm³, p= .09). Analysis by type of pneumonia demonstrated that the CD4 count was nearly identical among case and control patients in the PCP subgroup (0.041 vs 0.041 × 10⁹/mm³, p= .66) but somewhat lower among BP case patients than control patients (0.024 vs 0.044 × 10⁹/mm³, p= .052). A significantly greater proportion of BP case patients had CD4 counts less than the cutoff value of 0.050 × 10⁹/mm³.

The CD4 counts for this study population were uniformly low and therefore of little predictive value in distinguishing case from control patients. However, prior hospitalization within the previous 6 months and number of concurrent AIDS diagnoses, two other measures of disease progression, were both risk factors for early readmission. Coincident AIDS diagnoses included Kaposi's sarcoma, candidiasis, mycobacterial disease, cytomegalovirus infection, central nervous system lymphoma, cryptococcus, toxoplasma, histoplasma, or coccidiomycosis infection, and AIDS dementia complex. When analyzed by pneumonia type, these two predictor variables were only significant in the BP subgroup (Table 2). History of previous admissions for PCP or BP, previous pneumothoraces, and history of psychiatric illness were not predictive of early readmission.

Multivariate Analysis

We created multivariate regression models for the overall group and for each subgroup using the following predictor variables: leaving the hospital alone, crack cocaine use, one or more coincident AIDS diagnoses, hospitalization in the 6 months preceding the index hospitalization, and CD4 count less than 0.050 × 10⁹/mm³(Table 4) In the combined analysis, the four risk factors revealed in the univariate analysis remained independently significant. In the PCP subgroup, leaving the hospital alone remained the single strongest risk factor for early readmission. In the BP subgroup, coincident AIDS diagnoses and recent previous admission remained strong predictors, while CD4 × 10⁹/mm³did not. Leaving the hospital alone was a positive but nonsignificant univariate predictor in the smaller BP subgroup but a significant multivariate predictor (Table 4). These models were recreated without the CD4 variable with no significant change in the combined or PCP group results.

Table 4.

Predictors of Early Readmission: Multivariate Analysis by Pneumonia Type ^*

	PCP	Bacterial Pneumonia	Combined
Left hospital alone	5.97 (2.09, 17.1; .0009) †	12.4 (1.20, 128.4; .034) †	4.76 (2.06, 11.0; .0003) †
Crack use	2.74 (0.60, 12.4; .19)	15.1 (0.7, 312.5; .079)	3.40 (1.02, 11.3; .046) †
>1 Coincident AIDS diagnosis	2.07 (0.70, 6.20; .19)	46.6 (2.0, 999.0; .017) †	3.65 (1.44, 9.26; .0065) †
CD4 < 50	0.50 (0.16, 1.53; .22)	12.8 (0.78, 211.4; .074)	0.82 (0.33, 2.06; .68)
Previous admission <6 months	1.98 (0.66, 5.95; .22)	23.6 (2.20, 251.7; .0089) †	2.82 (1.21, 6.57; .016) †

^*Values are given as odds ratios (95% confidence intervals; p values). PCP indicates Pneumocystis carinii pneumonia.

^†p< .05, by Fisher's Exact Test.

Reasons for Readmission

Although there were many individual reasons for readmission, several general categories were identified. Of the 90 case patients, 35 (39%) were readmitted with a subsequent unrelated diagnosis. The most common diagnoses were AIDS-dementia complex (6), cryptococcal infection (3), and progression of pulmonary Kaposi's sarcoma (2). Twenty-six (29%) of early readmissions were due to medical noncompliance or factors related to drug abuse or psychiatric symptoms. Remarkably, 21 (81%) of these 26 early readmissions occurred in patients who had left the hospital alone. Thirteen patients had illness that progressed despite appropriate therapy (9 PCP patients, including 4 with pneumothoraces, and 4 BP patients). Seven patients failed to thrive at home, requiring readmission for rehydration or placement into a more structured living environment or both. Finally, two patients with the clinical diagnosis of BP (including one patient who refused sputum induction) were readmitted and ultimately diagnosed with PCP (Table 5).

Table 5.

Reasons for Readmission for Case and Control Patients

Reason	Case Patients, n(%) (n= 90)	Control Patients, n(%) (n= 87)
Unrelated diagnosis	35 (39)	43 (73)
Noncompliance/drug use/psychotic behavior	26 (29)	0
Progression despite therapy	13 (14)	0
Incorrect diagnosis	2 (2)	0
Failure to thrive at home	7 (8)	3 (3)
Died at home or hospice	0	3 (3)
New episode of PCP/BP *	0	10 (11)
Not readmitted during study, or diagnosis not recorded	0	28 (32)

^*PCP indicates Pneumocystis carinii pneumonia; BP, bacterial pneumonia.

Reason for readmission was known for 59 of the 87 control patients (9 patients were not readmitted during the study period, 10 others were known to have left the study area at some point after the 2-week early readmission period, and 6 did not have readmission information recorded). Forty-three (73%) of the 59 were readmitted for unrelated illness. The most common were lymphoma (7), bacterial sepsis (6), cytomegaloviral retinitis (5), PCP (4, in those patients initially admitted with BP), and pulmonary Kaposi's sarcoma (3). Six patients had recurrence of PCP and 4 had recurrence of BP. The average time to recurrence in these 10 control patients was 16.2 weeks. Of the remaining 6 patients, 3 were readmitted for rehydration and 3 died at home or in hospice (Table 5).

DISCUSSION

Four variables known during the index hospitalization were significant risk factors for early readmission. Two were related to social factors: leaving the hospital alone and smoking crack cocaine. Two were related to burden of disease: concurrent AIDS diagnoses and hospital admission in the 6 months preceding the index hospitalization. Of note, neither clinical information describing the acute illness during the index hospitalization nor other forms of drug abuse were found to be predictive of early readmission.

Compared with studies of early readmission in other populations, the findings in our study are unique. However, we do observe some parallel with elderly populations, who are at increased risk of early readmission from widowhood and coping difficulty (insufficient home support systems, increased nursing needs, and poorer functional mobility).^11–13 We hypothesize that leaving the hospital alone at time of discharge, a finding distinct to this study, represents a sensitive marker of social isolation. The lack of significance of living situation, VNA home visits, designation of primary care physician, type of insurance, and marital status in predicting early readmission implies that leaving the hospital alone represents a readmission risk that is not directly related to access to care or level of home care. Leaving the hospital alone, a robust finding in the combined group and for each pneumonia subgroup, is clearly a surrogate marker for some, as yet undefined, underlying social deficit.

Crack use as a risk factor for readmission is probably unique to the indigent populations served by county hospitals such as SFGH, and it serves as a good example of the need to study marginalized populations that often differ in many ways from general medical populations. Other harmful drug use, such as alcoholism, tobacco smoking, and intravenous heroin or amphetamine use, were not risk factors in this study. Although there are data on the effect of injection drug use on AIDS progression,^16,17 it is not known whether crack use has a more detrimental effect on the health of patients with AIDS. The “culture” of crack use may be more socially isolating and, as with leaving the hospital alone, may represent a marker of some underlying social deficit that leaves patients at risk of early readmission. Alternatively, the power to find that other types of drug use are risk factors may have been limited by the small numbers in this study.

The patients in this study uniformly had low CD4 counts. Having a second AIDS-related illness (coincident AIDS diagnosis) and having been hospitalized within the past 6 months most likely represent markers of further disease progression that distinguish a “healthier” patient with a very low CD4 count from a “sicker” patient with the same CD4 count. In future research, it will be interesting to determine whether these findings correlate with HIV viral load levels.

A significant proportion (29%) of the case patients were readmitted for reasons related to personal behavior (e.g., noncompliance, illicit drug use, and psychiatric symptoms), outweighing the readmissions that could possibly be attributed to patients' burden of disease (progression despite therapy; 14%) or to deficiencies in medical care discharge planning (incorrect diagnosis, failure to thrive at home, 10%). Moreover, 21 (81%) of the 26 case patients readmitted for reasons related to noncompliance, active drug use, or psychiatric symptoms had initially left the hospital alone. In contrast, for the control patients, almost all of the known readmissions were due to a new episode of illness unrelated to the index admission (90%).

We conclude from these results that decreasing the rate of early readmission, a financial burden on our health care system and a personal burden on patients and their families and friends, requires a better understanding of the social milieu that defines indigent AIDS patients. Many have argued that with the increased emphasis on shorter hospital stays, patients would benefit from more intensive case management. However, studies on the effect of case management have yielded conflicting results. Although some studies have shown a reduction in days hospitalized and total admissions,^18,19 others have found no benefit,^20,21 including one study that specifically assessed case management of AIDS patients.²²

This study is limited by a retrospective design and reliance on chart review. Thus, errors may arise in diagnosis codes, social history, and other factors gleaned from the charts. In order to define a medically homogeneous group of AIDS patients, we limited our study to patients admitted with PCP and BP. These two infections remain the most common cause of hospitalization and death among AIDS patients,^23–26 and they are relatively easily and consistently diagnosed. Moreover, optimal inpatient treatment methods for both diseases are standardized and widely accepted. These characteristics reduce the likelihood of incorrect diagnoses and nonstandard treatment plans, common problems in hospital readmission studies.

Given the frequency of repeated hospitalizations in AIDS patients (almost twice a year in one study ²⁷) we chose the narrow readmission window of 14 days for our analysis. In doing so, we were guided by the principle that most patients, no matter how sick on admission, should experience a reasonable period of medical stability after discharge. By reviewing clinic charts and subsequent SFGH admission notes, we attempted to exclude any patients who may have been admitted elsewhere. It is possible, however, that a control patient may have had an early readmission at another hospital that was not noted in subsequent SFGH clinic or admission notes. In addition, the retrospective design prevented us from determining the mode of leaving the hospital in those patients for whom this information was missing in the Nursing Discharge Summary. To overcome this limitation, future prospective studies should include direct patient interviews prior to discharge.

We have found that patients with poor social support, crack users, and those with advanced disease are more likely to be noncompliant, actively use drugs, have psychiatric symptoms, or progress despite therapy. With advances in AIDS pharmacotherapy, specifically highly active antiretroviral therapy, there has been a decline in AIDS morbidity and mortality and rates of hospitalization. Thus, the risk factors associated with burden of disease (coincident AIDS diagnoses and recent antecedent hospitalization) and readmissions related to progression despite medical therapy are likely to be attenuated in this new era. However, the problems associated with social isolation, illicit drug use, and nonadherence with therapy remain. Further efforts to reduce early readmission in this population, therefore, should include prospective research on the social structures of indigent AIDS patients (particularly crack users and patients with psychiatric disease) and on interventions that improve their adherence to medical therapy.