JAIDS Journal of Acquired Immune Deficiency Syndromes:
An Electronic Medical Record-Based Model to Predict 30-Day Risk of Readmission and Death Among HIV-Infected Inpatients
Nijhawan, Ank E. MD, MPH*,†; Clark, Christopher MPA*; Kaplan, Richard MD‡; Moore, Billy PhD*; Halm, Ethan A. MD, MPH†; Amarasingham, Ruben MD*,†
*Department of Medicine, Division of Infectious Diseases, University of Texas Southwestern Medical Center, Center for Clinical Innovation, Parkland Health and Hospital System, Dallas, TX
†Departments of Internal Medicine and Clinical Sciences, University of Texas Southwestern Medical Center, Dallas, TX
‡Texas Center for Integrative Medicine, Internal and Integrative Medicine, Dallas, TX.
Correspondence to: Ank E. Nijhawan, MD, MPH, Department of Medicine, Division of Infectious Diseases, University of Texas Southwestern Medical Center, Center for Clinical Innovation, Parkland Health and Hospital System, 5123 Harry Hines Blvd. Dallas, TX 75235 (e-mail: firstname.lastname@example.org).
Supported by Parkland Health and Hospital System. A.E. Nijhawan was funded in part by the American Sexually Transmitted Diseases Association Developmental Award.
The authors have no conflicts of interest to disclose.
Presented at the XVIII AIDS Conference, July 19, 2010,Vienna, Austria.
Received April 27, 2012
Accepted August 10, 2012
Background: Readmission after hospitalization is costly, time-consuming, and remains common among HIV-infected individuals. We sought to use data from the Electronic Medical Record (EMR) to create a clinical, robust, multivariable model for predicting readmission risk in hospitalized HIV-infected patients.
Methods: We extracted clinical and nonclinical data from the EMR of HIV-infected patients admitted to a large urban hospital between March 2006 and November 2008. These data were used to build automated predictive models for 30-day risk of readmission and death.
Results: We identified 2476 index admissions among HIV-infected inpatients who were 73% males, 57% African American, with a mean age of 43 years. One-quarter were readmitted, and 3% died within 30 days of discharge. Those with a primary diagnosis during the index admission of HIV/AIDS accounted for the largest proportion of readmissions (41%), followed by those initially admitted for other infections (10%) or for oncologic (6%), pulmonary (5%), gastrointestinal (4%), and renal (3%) causes. Factors associated with readmission risk include: AIDS defining illness, CD4 ≤ 92, laboratory abnormalities, insurance status, homelessness, distance from the hospital, and prior emergency department visits and hospitalizations (c = 0.72; 95% confidence interval: 0.70 to 0.75). The multivariable predictors of death were CD4 < 132, abnormal liver function tests, creatinine >1.66, and hematocrit <30.8 (c = 0.79; 95% confidence interval: 0.74 to 0.84) for death.
Conclusions: Readmission rates among HIV-infected patients were high. An automated model composed of factors accessible from the EMR in the first 48 hours of admission performed well in predicting the 30-day risk of readmission among HIV patients. Such a model could be used in real-time to identify HIV patients at highest risk so readmission prevention resources could be targeted most efficiently.
The introduction of highly active antiretroviral therapy (HAART) about 15 years ago has dramatically improved overall survival among persons infected with HIV.1–3 However, although admissions related to advanced AIDS and opportunistic infections have declined significantly in the post-HAART era, the rate of inpatient admission for any cause among HIV-infected individuals remains relatively high4–6 with a slow rate of decline.7 Women,5,7–10 African Americans,5,7,11 injection drug users,5,7,8,12–14 those with mental illness,13,15 those not receiving HAART,4,9 and those with low CD4 counts12,14 are significantly more likely to be admitted to the hospital. Readmission among HIV-infected individuals is common8,10,16 and is influenced by both medical and social factors.16–18
The high rate and cost of hospital readmission among the elderly has been the source of considerable research and policy interest.19 The greatest attention to date has been on efforts to reduce the risk of readmission among patients with congestive heart failure, pneumonia, and myocardial infarction—the 3 most common medical reasons for hospitalization among Medicare beneficiaries. Other costly conditions, such as HIV, have been less well studied and merit serious attention. Data from 6 states in 2004 found that a single HIV admission costs more than $15,000, with an average length of stay of 8.4 days, considerably longer than most hospitalizations in the non-HIV population.20,21 Efforts to better understand, predict, and reduce readmissions among patients infected with HIV have the potential to improve patient outcomes and decrease costs.
To prevent readmissions, a rapid and reliable method to assess and predict readmission risk is needed early in the index admission. Unfortunately, physicians, case managers, and nurses are not able to accurately predict readmissions using standard approaches.22 The best performing models of the risk of 30-day readmission have been disease specific and incorporate detailed data on both clinical and social factors.23–25 The increasing use of comprehensive electronic medical records (EMRs) provides the opportunity to develop sophisticated risk prediction rules harnessing the broad range of detailed information available electronically.
This study sought to (1) characterize the rates and reasons for readmission with 30 days of discharge among patients hospitalized with HIV, (2) develop and validate a model to predict 30-day risk of readmission, based on a diverse set of clinical and social factors ascertainable from the EMR in the first 48 hours of admission, and (3) characterize rates of 30-day rates of death and factors associated with mortality.
Study Population, Data Source, and Variables
The study focused on HIV-infected patients hospitalized on the internal medicine service between March 1, 2006 and November 30, 2008 at Parkland Memorial Hospital, a large urban teaching hospital in Dallas, TX. Parkland Health and Hospital System (Parkland) is an integrated delivery system, which also includes a network of 11 adult-oriented community outpatient clinics, including a focused HIV clinic. It is the sole public safety net provider in Dallas County. This study focuses on patients with a principal or secondary discharge diagnosis of HIV or AIDS (based on ICD-9 codes 042 and V08). All variables were extracted from the Epic EMR (EPIC Systems Corporation, Verona, WI) of the health system.
Outcome Variables—Readmission and Mortality
We developed 2 separate models among patients hospitalized with HIV, one for risk of readmission for any cause within 30 days of discharge and a different one predicting risk of death within 30 days of discharge. Readmission was defined as an inpatient admission for any cause to any hospital within a 30-day period from the index discharge. If a readmission occurred after 30 days of an HIV index admission discharge, it was considered as another index admission. For the mortality model, all index admissions were included. For the readmission model, we excluded all index admissions that resulted in death within 30 days of discharge.
We identified readmissions to any hospital in the area using a data linkage service available through the Dallas-Fort Worth Hospital Council that maintains a regional hospital discharge data set of all admissions to 70 acute care hospitals in North Texas.26 Survival in the 30 days after discharge was ascertained by querying the Parkland inpatient and outpatient clinic data systems for evidence of a follow-up encounter more than one month after the index hospitalization. For patients lacking an EMR encounter in the 30 days after discharge, we searched the Social Security Death Index.
We extracted a large number of candidate risk factors according to 3 criteria. The variable needed to be (1) available in the EMR, (2) routinely collected or available within the first 48 hours of hospital presentation, and (3) determined by consensus of an interdisciplinary team as reasonably available to most hospitals with a basic EMR.
We developed a conceptual framework of readmission based on the existing literature and clinical expertise on comorbidities and treatments.27 We identified 2 major categories of risk predictors, clinical and nonclinical variables. Clinical factors included HIV/AIDS severity, laboratory test results, mental health history, and HIV medication regimens. Nonclinical variables included sociodemographics, social instability (eg, number of home address changes in the prior year, homelessness, and lack of emergency contact information); health-risk behavior (eg, history of confirmed cocaine, opiate use during the past year using specific ICD-9 CM codes); adherence measures (eg, history of missed clinical appointments and leaving against medical advice); and prior acute health care utilization [hospitalizations and emergency department (ED) visits]. We also included a composite neighborhood measure of social “disadvantagedness” based on median household income, percent of households below the poverty level in the census tract, percent of residents with age >18 years, who reported with less than college or higher education, and percent of population who were non-white. The data were reported by the US Census Bureau for the 2000 census for 1046 census tracts in the 12-county area surrounding Dallas County. Census tract scores were used to rank the tracts into quintiles with the lowest one classified as a high-risk census tract.
Before model development, we examined all available variables for missing values. For analyses with continuous variables (age, length of stay, laboratory score, etc), we allowed case-wise deletion because we found missing values to be rare. For each categorical variable, we created a response level for grouping cases with missing values, allowing us to keep these cases in the analysis.
Derivation of the Electronic Readmission Models
The process of model building involved 4 stages. First, we examined the univariate relationship between the outcome variable readmission and each of the clinical and nonclinical categorical variables through logistic regression at threshold significance of P ≤ 0.20. We examined continuous variables (eg, age, number of prior ED visits, and inpatient admissions) for nonlinear effects by testing the contributions of spline functions and variable transformations.
We split each continuous laboratory variable into distinct categories using recursive partitioning to determine limit values (using RTREE).28 Second, to protect against excess overfitting, we restricted the number of predictor variables to that estimated through a heuristic shrinkage formula.29 Third, candidate variables were ranked by P value using bootstrapping in 1000 multivariate logistic regression iterations.30 In the fourth stage, we included variables that were selected based on conceptual and statistical significance to fit the final model. To account for the effects of patients with more than one index admission, we used robust variance–covariance matrix estimators for computing standard errors for model coefficients.31 The process was repeated for modeling mortality, resulting in a different set of covariates and coefficients.
The models were validated using several approaches. Model discrimination was assessed by the c statistic. Model calibration was evaluated using the Hosmer–Lemeshow χ2 goodness of fit test. We internally validated the model using a cross-validation methodology29,32 and stratified our cohort by risk quintiles. In this approach, the full sample was split randomly without replacement into derivation (50%) and validation (50%) subsamples 1000 times. The model was reconstructed for each of the derivation samples using stepwise regression and then tested on the corresponding validation subsample. The c statistics for the derivation and validation subsamples were then computed in each iteration and averaged across all iterations. Using interval endpoints determined by the derivation subsamples, 5 readmission risk categories (1 = very low to 5 = very high) were created based on quintiles of predicted risk and were graphically assessed by comparing derivation and validation subsample results. To assess the association between readmission time and the predicted risk, Kaplan–Meier curves were constructed for the stratified risk groups.
All analyses were conducted using STATA statistical software (version 10.0; STATA Corp, College Station, TX) and RTREE (from http://peace.med.yale.edu/pub). The University of Texas Southwestern Medical Center Institutional Review Board approved the research protocol.
Description of the Study Cohort
During the 32-month study period, there were 2476 index admissions among 1509 unique HIV-infected individuals. Two-thirds (1000, 66%) of individuals had only one index admissions, 291 (19%) had 2 index admissions, and the remaining individuals (218, 14%) had 3 index admissions. The characteristics of the study cohort are shown in Table 1. Three-quarters of admissions were among patients who were males (73%), more than half were African American (57%), and the mean age was 43 (SD = 9.4) years. More than half of admissions had a CD4 <200, and a quarter had an AIDS-defining illness within the year before admission. The average length of stay was 7 days.
TABLE 1-a Characteri...Image Tools
Predictors of 30-Day Risk of Readmission
TABLE 1-b Characteri...Image Tools
TABLE 1-c Characteri...Image Tools
Overall, one in 4 admissions resulted in a readmission within 30 days of discharge (25.2%, N = 623). More than 3-quarters of readmissions within 30 days were to the index hospital (77.8%). When the cohort is limited to those index admissions that did not result in death within 30 days of discharge (N = 2402), the readmission rate was similar (24.4%, N = 586).
The largest proportion of all-cause readmission were among those with a principal diagnosis at the index admission of HIV/AIDS (41%), followed by other infections (10%), cancer (6%), gastrointestinal bleeding (4%), congestive heart failure (3%), and renal failure (3%) (Table 2). The highest rates of readmissions were among those with an index admission for cancer, congestive heart failure, renal failure, and gastrointestinal disorders (Table 2).
In the univariate analyses, a large number of domains and variables were associated with risk of readmission (Table 3), including living situation, HIV/AIDS severity, mental health, substance use, laboratory abnormalities, insurance status, and prior ED visits/hospitalizations. Results of multivariate analysis for readmission are shown in Table 4. A broad range of clinical factors were associated with risk of readmission, including AIDS defining illness, CD4 ≤ 92, and renal, hepatic, and hematologic abnormalities. In addition, patients who were homeless, insured by Medicaid, and had higher rates of utilization of inpatient and emergency department services were more likely to be readmitted within 30 days of release. Linear shrinkage analysis did not reveal overfitting.
Predictors of 30-Day Risk of Death
TABLE 3-b Univariate...Image Tools
TABLE 3-c Univariate...Image Tools
The rate of death within 30 days of discharge was 3.0% (N = 74). A social security number was not available for 9% of patients and, therefore, 30-day mortality through the social security death index was unavailable for these subjects. Laboratory abnormalities, including renal failure, acidosis, liver abnormalities, anemia, and immunosuppression, rather than social factors, were the most important predictors of mortality in both univariate and multivariate analyses for mortality (Tables 3 and 5).
The electronic readmission model produced an overall c statistic of 0.72 [95% confidence interval (CI): 0.70 to 0.75]. The average c statistics from the cross-validation analysis were 0.72 (95% CI: 0.69 to 0.74) and 0.70 (95% CI: 0.67 to 0.72) for the derivation and validation samples, respectively.
The electronic model was capable of stratifying patients across a wide range of risk for readmission from 8% to 45%, and derivation and validation subsamples were highly concordant across the risk spectrum (Fig. 1). Patients in higher risk categories of the electronic model were readmitted earlier over the 30-day postdischarge period in a Kaplan–Meier analysis (P < 0.001, Fig. 2). For the electronic mortality model, the overall c statistic was 0.79 (95% CI: 0.74 to 0.84).
In this study of 2476 admissions among HIV-infected patients treated in a large urban safety net hospital, we found that 1 in 4 admissions resulted in a readmission within 30 days of discharge. Although there are limited published data on HIV readmission rates, 2 prior studies reported 14-day readmission rates among HIV patients with pneumonia of13%–19%.16,17 Our 30-day readmission rate is similarly high and is comparable to readmissions for congestive heart failure (21%–26%),24,33,34 and higher than published readmission rates for acute myocardial infarction (18.9%)35 and pneumonia (17.4%).36
We also developed an innovative EMR-derived multivariate model that accurately predicted 30-day readmission risk among HIV inpatients. This study extends an advancing field of work that successfully characterizes patient risk based on data available in the EMR.24,37–39 A broad range of clinical and nonclinical variables contributed to this prediction model, including HIV-related factors, such as CD4 count and a history of AIDS-defining illness; laboratory abnormalities, such as renal, hepatic, and hematologic abnormalities; and living situation, health insurance, and health care utilization.
Several studies have sought to characterize individual clinical and social factors that may increase the risk of readmission among HIV-infected patients. Palepu et al,16 in a retrospective case-control study, found that among HIV patients admitted with pneumonia, those who left against medical advice, those who lived in the poorest urban neighborhoods, and those who had been hospitalized in the past 6 months were at highest risk for readmission. Similarly, Grant et al,17 in a separate study of HIV patients admitted with pneumonia, found that patients who did not have a companion at discharge or who reported crack cocaine use were at higher risk for readmission. Noysk et al,18 in a propensity score matched analysis, found that taking HAART significantly reduced a patient's risk for readmission.
Our findings confirm and extend the data from these prior studies. We found that homelessness, living far from the hospital, Medicaid insurance, and frequent inpatient and emergency room health care utilization were important components in our prediction model. These results suggest that nonclinical factors may be contributing to readmissions, including housing instability, transportation, socioeconomic status and possibly poor engagement in outpatient care. Surprisingly, mental health issues and substance use, which were seen in 41% and 31% of inpatients respectively, and were correlated with readmission in univariate analyses, did not improve the performance of our multivariate prediction model. This was probably because other measures of social disadvantage were more important.
This study also provides data on the epidemiology of hospitalizations among patients with HIV. Not surprisingly, these patients were predominately male minorities with many social disadvantages and were severely immunocompromised. HIV/AIDS was the most common primary diagnosis at the time of index admission, and these patients accounted for a large proportion of readmissions (41%). Other infections were the next most common reason for readmission (10%). More than half of the patients had a CD4 count <200 cells/mm3 and 60% of these had a CD4 count <50 cells/mm3. Although other studies have shown that reasons for hospitalization among HIV patients has shifted from AIDS-defining illnesses to chronic end-organ diseases,6,12 our study emphasizes that AIDS is still a common cause of hospitalization. In addition, African Americans and Hispanics together comprise 74% of our cohort and nearly half of all patients were living in a census tract in the lowest socioeconomic quintile in Dallas. The overrepresentation of minorities and underserved individuals in the inpatient population highlights the severity of the HIV epidemic in Texas,40 especially among African Americans. Low CD4 cell counts and hospitalizations in this group may be partially explained by low rates of HAART utilization.11,41,42
Despite clinical indications for treatment, more than half of the patients in this study had no history of antiretrovirals (ARVs) being prescribed within the Parkland system, and approximately 25% of patients in this study did not fill their ARV prescription in the 30 days before admission. More than 20% had missed scheduled outpatient appointments in one of 11 clinics in the health system. To avoid complications due to HIV and related hospital admissions, patients need to adhere to both their ARVs and their outpatient clinic visits.43–45 Engagement in care, including both linkage to care and retention in care, has become an important focus among HIV providers and researchers, with implications for individual outcomes, including improved mortality,43,46,47 and population-level outcomes, such as reducing racial and socioeconomic disparities in health care and transmission of disease.43,48,49 To improve the high rates of 30-day readmission among HIV patients in our hospital system, recent efforts include providing clinic appointments within 2 weeks of discharge from the hospital and an ongoing in-depth chart review to identify preventable causes of readmission. Dallas County Health and Human Services also has a new outreach initiative to locate HIV-infected individuals who have been lost to care.
Several limitations are worth noting. Our study was based at a single urban safety net hospital serving predominantly poor minorities, so generalizability to other settings and patient populations is unknown. However, ours is the largest study on readmissions among HIV patients to date, and this population is one that carries the biggest burden of HIV. We had the advantage of being able to extract data from both inpatient and outpatient health information systems to create a richer and more longitudinal profile of patients and their risk factors. These results may not apply to hospitals lacking EMRs or those without any outpatient care data. Nonetheless, in considering potential predictors, we favored the selection of elements that would be reasonably available to hospitals with a basic EMR. The adoption of EMRs is accelerating; thus, automated models are likely to have wider application value in the near future. Another limitation is missing data, such as CD4 and viral load values, before admission. However, because our hospital system cares for most HIV patients in the county, we believe that this reflects a real-world setting where a large proportion of HIV-infected patients are not engaged in routine HIV care.50 Finally, although we used rigorous statistical methods to validate our model, our validation was limited to our internal data set. Future efforts should validate the model prospectively and optimally in other institutions and settings.
These findings have implications for both clinical care and health policy. Clinicians should recognize that HIV patients are at high risk for readmission and will need to more effectively engage patients, families, case management, and other services early during a hospitalization to address risk factors for readmission. From a policy perspective, this work highlights the powerful impact of both clinical and nonclinical factors on risk of readmission in HIV. Many of the social disadvantage and behavioral risk factors may be things that providers have little direct influence over and have nothing to do with the quality of inpatient or postdischarge care. However, these types of EMR-enabled disease-specific prediction models could prove useful in identifying those at highest risk who should be the focus of case management resources which are often quite limited in most safety net health systems. In addition, information on the specific things that put an individual at risk for readmission or death (clinical vs. social) could help facilitate a more patient-tailored approach to improving the transition of care.
Future research should explore the causes of readmission among HIV-infected patients in greater detail and determine what proportion of readmissions may be potentially preventable. These findings could inform future interventions, building upon prior successful linkage to care programs, such as the ARTAS case management intervention,51 based on promoting self-efficacy and project CONNECT,52 which involves client-oriented navigation. In addition, external and prospective validation of the EMR prediction model is needed, including using this model in real time. Once validated, the EMR model can be used to stratify hospitalized patients based on their readmission risk, thereby helping direct resources and interventions to those who may derive the greatest benefit.
The authors thank the Dallas-Fort Worth Hospital Council Foundation for assisting in the collection of posthospitalization data through the information sharing initiative.
1. Crum NF, Riffenburgh RH, Wegner S, et al.. Comparisons of causes of death and mortality rates among HIV-infected persons: analysis of the pre-, early, and late HAART (highly active antiretroviral therapy) eras. J Acquir Immune Defic Syndr. 2006;41:194–200.
2. Lima VD, Hogg RS, Harrigan PR, et al.. Continued improvement in survival among HIV-infected individuals with newer forms of highly active antiretroviral therapy. AIDS. 2007;21:685–692.
3. Palella FJ Jr, Baker RK, Moorman AC, et al.. Mortality in the highly active antiretroviral therapy era: changing causes of death and disease in the HIV outpatient study. J Acquir Immune Defic Syndr. 2006;43:27–34.
4. Crum-Cianflone NF, Grandits G, Echols S, et al.. Trends and causes of hospitalizations among HIV-infected persons during the late HAART era: what is the impact of CD4 counts and HAART use? J Acquir Immune Defic Syndr. 2010;54:248–257.
5. Gebo KA, Diener-West M, Moore RD. Hospitalization rates in an urban cohort after the introduction of highly active antiretroviral therapy. J Acquir Immune Defic Syndr. 2001;27:143–152.
6. Berry SA, Fleishman JA, Moore RD, et al.. Trends in Reasons for Hospitalization in a Multisite United States Cohort of Persons Living With HIV, 2001-2008. J Acquir Immune Defic Syndr. 2012;59:368–375.
7. Fleishman JA, Hellinger FH. Recent trends in HIV-related inpatient admissions 1996-2000: a 7-state study. J Acquir Immune Defic Syndr. 2003;34:102–110.
8. Fielden SJ, Rusch ML, Levy AR, et al.. Predicting hospitalization among HIV-infected antiretroviral naive patients starting HAART: determining clinical markers and exploring social pathways. AIDS Care. 2008;20:297–303.
9. Floris-Moore M, Lo Y, Klein RS, et al.. Gender and hospitalization patterns among HIV-infected drug users before and after the availability of highly active antiretroviral therapy. J Acquir Immune Defic Syndr. 2003;34:331–337.
10. Weber AE, Yip B, O'Shaughnessy MV, et al.. Determinants of hospital admission among HIV-positive people in British Columbia. CMAJ. 2000;162:783–786.
11. Oramasionwu CU, Hunter JM, Skinner J, et al.. Black race as a predictor of poor health outcomes among a national cohort of HIV/AIDS patients admitted to US hospitals: a cohort study. BMC Infect Dis. 2009;9:127.
12. Buchacz K, Baker RK, Moorman AC, et al.. Rates of hospitalizations and associated diagnoses in a large multisite cohort of HIV patients in the United States, 1994-2005. AIDS. 2008;22:1345–1354.
13. Himelhoch S, Chander G, Fleishman JA, et al.. Access to HAART and utilization of inpatient medical hospital services among HIV-infected patients with co-occurring serious mental illness and injection drug use. Gen Hosp Psychiatry. 2007;29:518–525.
14. Tashima KT, Hogan JW, Gardner LI, et al.. A longitudinal analysis of hospitalization and emergency department use among human immunodeficiency virus-infected women reporting protease inhibitor use. Clin Infect Dis. 2001;33:2055–2060.
15. Betz ME, Gebo KA, Barber E, et al.. Patterns of diagnoses in hospital admissions in a multistate cohort of HIV-positive adults in 2001. Med Care. 2005;43(suppl 9):III3–III14.
16. Palepu A, Sun H, Kuyper L, et al.. Predictors of early hospital readmission in HIV-infected patients with pneumonia. J Gen Intern Med. 2003;18:242–247.
17. Grant RW, Charlebois ED, Wachter RM. Risk factors for early hospital readmission in patients with AIDS and pneumonia. J Gen Int Med. 1999;14:531–536.
18. Nosyk B, Sun H, Li X, et al.. Highly active antiretroviral therapy and hospital readmission: comparison of a matched cohort. BMC Infect Dis. 2006;6:146.
19. Jencks SF, Williams MV, Coleman EA. Rehospitalizations among patients in the Medicare fee-for-service program. N Engl J Med. 2009;360:1418–1428.
20. Hellinger FJ. HIV patients in the HCUP database: a study of hospital utilization and costs. Inquiry. 2004;41:95–105.
21. Hellinger FJ. The changing pattern of hospital care for persons living with HIV: 2000 through 2004. J Acquir Immune Defic Syndr. 2007;45:239–246.
22. Allaudeen N, Schnipper JL, Orav EJ, et al.. Inability of providers to predict unplanned readmissions. J Gen Int Med. 2011;26:771–776.
23. Coleman EA, Min SJ, Chomiak A, et al.. Posthospital care transitions: patterns, complications, and risk identification. Health Serv Res. 2004;39:1449–1465.
24. Amarasingham R, Moore BJ, Tabak YP, et al.. An automated model to identify heart failure patients at risk for 30-day readmission or death using electronic medical record data. Med Care. 2010;48:981–988.
25. Kansagara D, Englander H, Salanitro A, et al.. Risk prediction models for hospital readmission: a systematic review. JAMA. 2011;306:1688–1698.
26. Dallas-Fort Worth Hospital C. Dallas-Fort Worth Hospital Council, Data Services. Irving, TX: DFWHC; 2011. Available at: http://www.dfwhc.org/about.html
Accessed March 2, 2012.
27. Arbaje AI, Wolff JL, Yu Q, et al.. Postdischarge environmental and socioeconomic factors and the likelihood of early hospital readmission among community-dwelling Medicare beneficiaries. Gerontologist. 2008;48:495–504.
28. Zhang HP, Singer B. Recursive Partitioning in the Health Sciences. New York, NY: Springer; 1999.
29. Harrell FE Jr, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15:361–387.
30. Efron B, Tibshirani R. An Introduction to the Bootstrap. New York, NY: Chapman and Hall/CRC Press; 1998.
31. Wei L, Lin D. Regression analysis of multivariate incomplete failure time data by modeling marginal distributions. J Am Stat Assoc. 1989;84:1065–1073.
32. Van Houwelingen JC, Le Cessie S. Predictive value of statistical models. Stat Med. 1990;9:1303–1325.
33. Krumholz HM, Parent EM, Tu N, et al.. Readmission after hospitalization for congestive heart failure among medicare beneficiaries. Arch Intern Med. 1997;157:99–104.
34. Hernandez AF, Greiner MA, Fonarow GC, et al.. Relationship between early physician follow-up and 30-day readmission among Medicare beneficiaries hospitalized for heart failure. JAMA. 2010;303:1716–1722.
35. Krumholz HM, Lin Z, Drye EE, et al.. An administrative claims measure suitable for profiling hospital performance based on 30-day all-cause readmission rates among patients with acute myocardial infarction. Circ Cardiovasc Qual Outcomes. 2011;4:243–252.
36. Lindenauer PK, Normand SL, Drye EE, et al.. Development, validation, and results of a measure of 30-day readmission following hospitalization for pneumonia. J Hosp Med. 2011;6:142–150.
37. Escobar GJ, Greene JD, Scheirer P, et al.. Risk-adjusting hospital inpatient mortality using automated inpatient, outpatient, and laboratory databases. Med Care. 2008;46:232–239.
38. Pine M, Jordan HS, Elixhauser A, et al.. Enhancement of claims data to improve risk adjustment of hospital mortality. JAMA. 2007;297:71–76.
39. Tabak YP, Johannes RS, Silber JH. Using automated clinical data for risk adjustment: development and validation of six disease-specific mortality predictive models for pay-for-performance. Med Care. 2007;45:789–805.
41. Lemly DC, Shepherd BE, Hulgan T, et al.. Race and sex differences in antiretroviral therapy use and mortality among HIV-infected persons in care. J Infect Dis. 2009;199:991–998.
42. Meditz AL, MaWhinney S, Allshouse A, et al.. Sex, race, and geographic region influence clinical outcomes following primary HIV-1 infection. J Infect Dis. 2011;203:442–451.
43. Giordano TP, Gifford AL, White AC Jr, et al.. Retention in care: a challenge to survival with HIV infection. Clin Infect Dis. 2007;44:1493–1499.
44. Mugavero MJ, Lin HY, Willig JH, et al.. Missed visits and mortality among patients establishing initial outpatient HIV treatment. Clin Infect Dis. 2009;48:248–256.
45. Ulett KB, Willig JH, Lin HY, et al.. The therapeutic implications of timely linkage and early retention in HIV care. AIDS Patient Care STDS. 2009;23:41–49.
46. Robbins GK, Daniels B, Zheng H, et al.. Predictors of antiretroviral treatment failure in an urban HIV clinic. J Acquir Immune Defic Syndr. 2007;44:30–37.
47. Keruly JC, Conviser R, Moore RD. Association of medical insurance and other factors with receipt of antiretroviral therapy. Am J Public Health. 2002;92:852–857.
48. Marks G, Crepaz N, Janssen RS. Estimating sexual transmission of HIV from persons aware and unaware that they are infected with the virus in the USA. AIDS. 2006;20:1447–1450.
49. Cohen MS, Gay C, Kashuba AD, et al.. Narrative review: antiretroviral therapy to prevent the sexual transmission of HIV-1. Ann Int Med. 2007;146:591–601.
50. Gardner EM, McLees MP, Steiner JF, et al.. The spectrum of engagement in HIV care and its relevance to test-and-treat strategies for prevention of HIV infection. Clin Infect Dis. 2011;52:793–800.
51. Gardner LI, Metsch LR, Anderson-Mahoney P, et al.. Efficacy of a brief case management intervention to link recently diagnosed HIV-infected persons to care. AIDS. 2005;19:423–431.
52. Mugavero MJ. Improving engagement in HIV care: what can we do? Top HIV Med. 2008;16:156–161.
readmission; electronic medical record; HIV/AIDS; health disparities; prediction model
© 2012 Lippincott Williams & Wilkins, Inc.
What does "Remember me" mean?
By checking this box, you'll stay logged in until you logout. You'll get easier access to your articles, collections,
media, and all your other content, even if you close your browser or shut down your
To protect your most sensitive data and activities (like changing your password),
we'll ask you to re-enter your password when you access these services.
What if I'm on a computer that I share with others?
If you're using a public computer or you share this computer with others, we recommend
that you uncheck the "Remember me" box.
Highlight selected keywords in the article text.
Data is temporarily unavailable. Please try again soon.