Influence of Frailty-Related Diagnoses, High-Risk Prescribing in Elderly Adults, and Primary Care Use on Readmissions in Fewer than 30 Days for Veterans Aged 65 and Older Jacqueline A. Pugh, MD,*†‡ Chen-Pin Wang, PhD,*§ Sara E. Espinoza, MD, MSc,†k# Polly H. Noe¨l, PhD,*k Mary Bollinger, PhD,*§ Megan Amuan, MPH,** Erin Finley, PhD, MPH,*k and Mary Jo Pugh, PhD*§

OBJECTIVES: To determine the effect of two variables not previously studied in the readmissions literature (frailty-related diagnoses and high-risk medications in the elderly (HRME)) and one understudied variable (volume of primary care visits in the prior year). DESIGN: Retrospective cohort study using data from a study designed to examine outcomes associated with inappropriate prescribing in elderly adults. SETTING: All Veterans Affairs (VA) facilities with acute inpatient beds in fiscal year 2006 (FY06). PARTICIPANTS: All veterans aged 65 and older by October 1, 2005, who received VA care at least once per year between October 1, 2004, and September 30, 2006, and were hospitalized at least once during FY06 on a medical or surgical unit. MEASUREMENTS: A generalized linear interactive risk prediction model included demographic and clinical characteristics (mental health and chronic medical conditions, frailty-related diagnoses, number of medications) in FY05; incident HRME in FY06 before index hospitalization or readmission; chronic HRME in FY05; and FY05 emergency department (ED), hospital, geriatric, palliative, or primary care use. Facility-level variables were complexity, rural versus urban, and FY06 admission rate. RESULTS: The mean adjusted readmission rate was 18.3%. The new frailty-related diagnoses variable is a risk

From the *Veterans Evidence-Based Research, Dissemination, and Implementation Center, †Geriatric Research Education and Clinical Center, South Texas Veterans Health Care System, ‡Division of Hospital Medicine, §Department of Epidemiology and Biostatistics, kDivision of Clinical Epidemiology, #Division of Geriatrics, Gerontology and Palliative Medicine, Department of Medicine, University of Texas Health Science Center at San Antonio, San Antonio, Texas; and **Center for Health Quality, Outcomes and Economic Research, Edith Nourse Rogers Memorial Hospital, Bedford, Massachusetts. Address correspondence to Jacqueline Pugh, VERDICT 11C-6, STVHCS, 7400 Merton Minter Blvd., San Antonio, TX 78229. E-mail: [email protected] DOI: 10.1111/jgs.12656

factor for readmission in addition to Charlson comorbidity score. Incident HRME use was associated with lower rates of readmission, as were higher numbers of primary care visits in the prior year. CONCLUSION: Frailty-related diagnoses may help to target individuals at higher risk of readmission to receive more-intensive care transition services. HRME use does not help in this targeting. A higher number of face-to-face primary care visits in the prior year, unlike ED and hospital use, correlates with fewer readmissions and may be another avenue for targeting prevention strategies. J Am Geriatr Soc 62:291–298, 2014.

Key words: early readmissions; high-risk prescribing in the elderly; frailty; primary care

E

arly readmissions (readmissions within 30 days of a previous hospital admission) are frequent, affecting one-fifth of Medicare beneficiaries hospitalized in 2003–04,1 and have seen little decline through 2011 despite considerable effort at improving care transitions and after care.2 The Patient Protection and Affordable Care Act designates reduction of avoidable rehospitalizations, long thought to be a marker of quality of care,3 as one of its methods of reducing healthcare costs while improving quality of care, with a goal of 20% reduction by the end of 2013.4 The cost of avoidable readmissions for Medicare is estimated to be $12 billion a year.5 Despite the policy stance already taken, the extent of preventability is unclear, with estimates in the literature ranging from 5% to 79%,6 at least in part because of varying definitions of avoidable or preventable.6,7 A recent metaanalysis of 16 studies7 found that, overall, 23% of readmissions were avoidable but that the proportion varied significantly (5–58.6%) according to study, number of chart reviewers, and teaching status of the hospital.

JAGS 62:291–298, 2014 Published 2014. This article is a U.S. Government work and is in the public domain in the U.S.A.

0002-8614/14/$15.00

292

PUGH ET AL.

Many studies including multiple risk-prediction models have attempted to identify risk factors for those at highest risk of readmission.8,9 Commonly studied risk factors have included specific medical diagnoses or comorbidity indices, mental health comorbidities (mental illness, alcohol or substance use), illness severity (severity index, laboratory findings, other), prior use of medical services (hospitalizations, emergency department (ED) visits, clinic visits or missed visits, index hospital length of stay), overall health and function (functional status, activity of daily living dependence, mobility, self-rated health, quality of life, cognitive impairment, visual or hearing impairment), sociodemographic factors (age, sex, race or ethnicity), and social determinants of health (socioeconomic status, income and employment status, health insurance status, education, marital status and number of people in the home, caregiver availability or other support, access to care or limited access (e.g., rural area), discharge location (home, nursing home)).8,9 Despite this plethora of data on risk factors, prediction models are only modestly better than chance at predicting readmission,8 and transition-of-care interventions have been shown to have limited success.10 Learning to better identify who is truly at risk, as well as more about what usual (and potentially modifiable) medical care may be associated with readmissions, may help reduce readmissions by targeting resources at those at highest risk. After careful review of this literature, this study sought to identify risk factors currently available in medical databases (thereby not requiring extra patient data collection), not yet widely studied in the readmissions literature, and that might lead to more-efficient targeting of transitions-of-care interventions or suggest changes to primary care practices to prevent the index admission and readmission. For purposes of better targeting of resources, it was proposed that frailty might be a stronger predictor of readmission than the more-general comorbidity indices used. Frailty has been associated with risk of hospitalization.11,12 Most medical databases do not have a direct measurement of frailty; frailty-related diagnoses were chosen if they have been used as a major frailty characteristic (e.g., involuntary weight loss) or if they have been associated with frailty in prior studies that used the Fried model of frailty.11 Markers chosen were coagulopathy,13 involuntary weight loss,11 fluid and electrolyte imbalance,14 anemia,15 and fall or fracture.11 For purposes of improving care before admission or readmission, care practices that were potentially amenable to change were looked for. Previous work has shown that inappropriate prescribing in elderly adults is a risk factor for admission,16–18 but such has not yet been shown for readmission. One study found that elderly intensive care unit survivors were at high risk of being discharged on inappropriate medications,19 suggesting a potential area for intervention. Greater healthcare use before admission predicts readmission, but most studies have examined admissions or ED visits, with many fewer exploring clinic visits or missed clinic visits.8,20–24 The current study specifically explored the effect of number of primary care visits in the prior year on readmission. Many readmissions are in individuals with end-stage or terminal disease. A few studies have examined the effect of palliative care on readmission, with mixed results.25,26 The current study sought

FEBRUARY 2014–VOL. 62, NO. 2

JAGS

to add to this literature. It was possible to identify geriatric outpatient visits in the database, so the effect of these visits was specifically explored as well.

METHODS Study Design, Setting, and Sample This was a retrospective cohort study using data from a study designed to examine outcomes associated with inappropriate prescribing in elderly adults. The cohort included veterans aged 65 and older by October 1, 2005 (beginning of fiscal year 2006 (FY06)) and who received Veterans Affairs (VA) care at least once per year between October 1, 2004, and September 30, 2006. Those who had at least one medical or surgical hospital admission between October 1, 2005, and September 30, 2006, were included in this analysis (n = 134,839); individuals whose index admissions were to psychiatry were excluded from these analyses, as were individuals who died during their index admission. Individuals who died within 30 days of discharge from their index admission were excluded from the main model but included in a second model (Online Appendix S1) to understand their influence on the primary outcome of interest: readmission. Admissions with a primary diagnosis of cancer were also excluded to avoid counting scheduled readmissions for chemotherapy. Institutional review boards at the University of Texas Health Science Center at San Antonio, the Hines VA, and the Bedford VA approved this study.

Data Sources National VA inpatient, outpatient, and pharmacy data from October 1, 2004, through October 31, 2006, were obtained for individuals aged 65 and older at the beginning of October 1, 2005. A merged database was created using inpatient and outpatient records from the VA National Patient Care Database and all outpatient pharmacy prescription data from the VA Pharmacy Benefits Management database. Records were merged using an encrypted identifier that is consistent for each person across VA data sets.

Measures Readmission within 30 days was defined as any admission 24 hours or more and less than 30 days after discharge from an index admission in FY06. An index admission was defined as the first of multiple admissions that occur within 30 days of each other. Readmission rates for facilities were calculated based on whether any readmission (≥1) occurred for an individual patient at that facility; 91.8% (n = 20,306) of individuals had one readmission, 7.4% (n = 1,654) had two, and 0.62% (n = 157) had three or four.

Primary Independent Variables Frailty-Related Diagnoses Five frailty-related diagnoses were identified based on review of the literature and expert clinical opinion, as

JAGS

FEBRUARY 2014–VOL. 62, NO. 2

explained in the introduction. Markers chosen were coagulopathy,13 involuntary weight loss,11 fluid and electrolyte imbalance,14 anemia,15 and fall or fracture.11 This measure was created using algorithms from the Elixhauser comorbidity index27 except for falls, for which the Healthcare Effectiveness Data and Information Set (HEDIS) drug–disease interaction measure was used.28 The frailtyrelated diagnoses measure was dichotomized (0 vs ≥1).

High-Risk Medication in the Elderly Exposure Use of any of the HEDIS High-Risk Medication in the Elderly (HRME) drugs was identified using VA pharmacy data in FY05 for chronic HRME use and FY06 before the index admission or between the index discharge date and the readmission or 30 days after discharge if there was no readmission for incident HRME use.29 Individuals who had HRME use in FY05 and an additional new HMRE prescription in FY06 were counted as chronic HRME only. The list of HRME can be found in a previous publication29 and in Online Appendix S2.

Primary Care Better access to primary care has been associated with fewer admissions for ambulatory care–sensitive conditions, so number of visits to primary care in the previous fiscal year was included, categorized into three levels (0–1, 2–4, ≥5).

Control Variables Patient demographic characteristics (age, sex, race and ethnicity) were identified using data fields from VA administrative databases between FY03 (October 1, 2002, to September 30, 2003) and FY06 (October 1, 2005, to September 30, 2006). Poverty was defined using the “exempt from copayment” variable. Level of service-connected disability and income means testing determine priority status in the Veterans Health Administration (VHA). Veterans may be classified into one of eight priority categories. These categories determine whether a veteran pays a copayment for prescriptions or receives free VA health care. Veterans with incomes below the basic pension level of $9,556 in 2002 paid no copayments.30 Priority status can thus be used to assess potential financial barriers to care by comparing those who do and do not pay copayments for care. Previous research on veterans has used priority status as a proxy for socioeconomic status and severity of disease, with a greater disease burden being associated with lower socioeconomic status in veterans.31,32 Health status variables in the analyses included several measures of disease burden. Because previous studies of predictors of readmission identified physical comorbidities and mental illness, these were also included as covariates.8 International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes found in VA inpatient and outpatient data (diagnoses in two outpatient or one inpatient encounter) (FY05 only) were used to identify individuals with physical and psychiatric conditions. The following psychiatric conditions included in the Selim Psychiatric Comorbidity Index were identified: schizophrenia, bipolar disorder, depressive disorder, posttraumatic

FRAILTY, PRESCRIBING, PRIMARY CARE, AND READMISSIONS

293

stress disorder, substance use disorder, and anxiety disorders.33,34 Because of the highly skewed distribution, the psychiatric comorbidity measure was dichotomized (0 vs ≥1). For physical comorbidity, a continuous variable for a comorbidity index was created based on the methods of Charlson and colleagues35 and adapted by Deyo and colleagues,36 creating a score (range 0–34) based on the presence of 18 chronic conditions documented in the electronic medical record using ICD-9 codes. Higher scores denote greater levels of comorbidity.

Measures of Geriatric Care and Palliative Care Use in FY05 Because some models of geriatric care are associated with better quality of care, better quality of life, improved function, and less healthcare use,37 a variable for whether the individual had been seen in a geriatric clinic in FY05 was included.38 Because palliative care has been shown to reduce readmissions,26 whether the patient had received palliative care in FY05 was included. Because the literature on readmissions has shown that prior use of EDs and hospitalization is a strong predictor of future use,9 individuals with ED or hospital care in FY05 were identified. The number of unique medications (not including supplies or vitamins that are included in the pharmacy files but including acute prescriptions such as antibiotics) each individual received during FY05 (October 1, 2004, to September 30, 2005) was counted to assess polypharmacy. Counts of all medications prescribed were included and categorized as 0 to 5, 6 to 8, 9 to 11, and 12 and more.

Facility-Level Variables VHA categorizes facilities according to complexity level based on the characteristics of the patient population, clinical services offered, educational and research missions, and administrative complexity. Facilities are classified into three levels, with Level 1 representing the most-complex facilities, Level 2 moderately complex facilities, and Level 3 the least-complex facilities. Level 1 is further subdivided into categories a to c.39 The model variables, definitions, and weightings are described elsewhere.40 A facility rate of admissions was calculated for this population of facilities for FY06 as a marker of customary admitting rates, given recently published Medicare data suggesting this as a prominent predictor of readmissions.41 The facility from which an individual received the majority of his or her care in FY05 to FY06 was classified as urban or rural based on the VHA schema that designates locations throughout the United States as urban, rural, or highly rural.42

Analytical Strategy Chi-square tests were used to compare each categorical independent variable and two-sample t-tests to compare each continuous independent variable of individuals with no readmission with those of individuals with one or more readmissions. The generalized linear model (GLIM),43 assuming a binomial distribution, was used to test for the effect of individual- and facility-level variables on the

294

PUGH ET AL.

log-odds of 30-day readmission. Individual-level predictors included demographic characteristics (exempt from copayment, age, race and ethnicity, sex), medication exposure (chronic and incident HRME exposure and polypharmacy), comorbidities (psychiatric, medical, and frailty-related diagnoses), and prior use (any ED or hospitalization in FY05, number of primary care visits, geriatric clinic visits, palliative care in FY05). Facility-level predictors included complexity, rate of admissions, and location (urban vs rural). A random effect was posited in GLIM to account for within-facility correlation and additional or unobserved facility-level variation associated with readmission. Proc GLIMMIX Procedure in SAS 9.2 (SAS Institute, Inc., Cary, NC) was used to perform GLIM analyses. Additional analyses included a GLIM model including individuals who died within 30 days of the date of discharge from the index admission and a GLIM model including interaction terms between frailty-related diagnoses and HRME.

RESULTS One hundred twenty-six VA facilities had the complexity variable available and therefore were included in the analyses. In FY06, 4,662 individuals aged 65 and older were admitted to medical or surgical units and died during their first admission and so were excluded from all of the analyses; 5,755 died within 30 days of discharge from their index hospitalization but were included in the analyses. (See Online Appendix S3 for flowchart of patients and Online Appendix S1 for the results of the model including those who died.) One hundred twenty-nine thousand four hundred survived at least 30 days after their first medical or surgical hospitalization in FY06; 23,950 (18.5%) had at least one readmission within 30 days. Although 91.8% of the individuals with a readmission had only one readmission, 0.62% had three or four after different index admissions. A few individuals had more than one readmission within 30 days of a single index admission. Adjusted rates of readmission varied according to facility, from a low of 10.6% to a high of 25.1% (median 18.3%, interquartile range 16.8–19.8%). Table 1 displays demographic, comorbidity, medication usage and exposure, and healthcare use data for individuals with no readmission and those with one or more readmissions. Those who were readmitted had a significantly greater likelihood of being older, African American or Caucasian (as opposed to Hispanic or race missing), male, poor (exempt from copayment), taking 12 or more medications, at a higher level of comorbidity, or with one or more of the five frailty-related diagnoses (coagulopathy, involuntary weight loss, fluid and electrolyte imbalance, anemia, fall or fracture). Those with readmissions were less likely to have incident HRME exposure and slightly more likely to have chronic HRME exposure. There was no difference between the two groups in percentage with mental illness. With regard to use, individuals with readmissions had a higher likelihood of a previous hospitalization or ED visit or being seen in a geriatrics clinic or in palliative care in FY05. Those who were readmitted were less likely to have the highest usage of primary care but also less likely to have the lowest level of usage (0–1). Readmitted individuals were more likely to be from rural

FEBRUARY 2014–VOL. 62, NO. 2

JAGS

and less-complex facilities. Consistent with the literature, a higher admission rate in a facility was associated with a significantly higher proportion of individuals with readmission (not shown on Table 1). Next the contribution of the various individual-, provider-, and facility-level variables in readmission rates were examined using GLIM analyses.43 As shown in Table 2, the new database-derived frailty-related diagnoses variable was associated with risk of readmission in addition to the effect of the Charlson Comorbidity Index on readmissions. Once frailty-related diagnoses are in the model, age is no longer a significant predictor of readmission in this cohort of individuals aged 65 and older. The primary care faceto-face visit frequency in FY05 was protective of readmission, with higher levels becoming more protective. The third independent variable, incident HRME exposure in FY06 before index hospitalization, or between discharge and readmission or 30 days after discharge, was associated with lower rates of readmission, not higher. Other variables in the model as shown on Table 2 were consistent with the readmissions literature except for mental illness. Having a diagnosis of mental illness was associated with a lower rate of readmission. Facility-level admission rate was associated with risk of readmission (odds ratio (OR) = 1.02 for every 1% increase in facility level admission rate, 95% confidence interval (CI) = 1.01– 1.03). Geriatric and palliative care exposure were not protective of readmission. The C-statistic for the model was 0.6554. A second model was run including all individuals who died within the 30 days after discharge from the index hospitalization. The results of this model can be found in an Online Appendix S1. Although the magnitude of significance changed a small amount for some of the variables, none changed direction or interpretation.

DISCUSSION Unique individual-level predictors of readmission assessed in this study include HRME and a new frailty marker derived from frailty-related diagnoses in administrative data. Frailty-related diagnoses were a modest predictor of readmission in the model, conveying a 15% greater risk than in those without any of these diagnoses; in comparison, the Charlson Comorbidity Index conveys a 13.7% greater risk for every 1-point increase in index. The modifiability of frailty progression is still being studied, and the frailty-related diagnoses used in the current study have not been compared directly with the physical assessment of frailty. These frailty-related diagnoses may function as a marker of need for more-intense intervention to prevent hospitalization. Because frailty is typically measured using physical performance measures and administered questionnaires, it was not possible to measure frailty directly using the administrative data set. Therefore, diagnoses that have been associated with frailty from prior studies were used. Although other definitions for identifying frailty exist,44,45 it was decided to use the Fried model because it has been the most widely applied in population-based studies; it has been validated in several cohort studies,12,46,47 including the original cohort (the Cardiovascular Health Study) in which the criteria were developed and shown to be

JAGS

FEBRUARY 2014–VOL. 62, NO. 2

FRAILTY, PRESCRIBING, PRIMARY CARE, AND READMISSIONS

295

Table 1. Participant Characteristics (n = 129,400) Characteristic

Demographic, % Age 65–74 75–84 ≥85 Race and ethnicity Black Hispanic Caucasian Other Missing Male Exempt from copayments, % Comorbidity and medication exposure, % High-risk medication exposure None Incident Chronic Number of medications 0–5 6–8 9–11 ≥12 Any mental illness diagnosis Weighted Charlson Comorbidity Index, mean  SD Frailty-related diagnoses, % Utilization, % Number of primary care visits in 2005 0–1 2–4 ≥5 Any geriatric clinic visit in 2005 Any emergency department visit in 2005 Any hospitalization in 2005 Palliative care in FY05 Facility characteristics, % Urban location Complexity level 1a 1b 1c 2 3

n

No Readmission, n = 105,450

30-Day Readmission, n = 23,950

P-Value

59,145 57,167 13,088

46.1 43.9 10.0

44.1 45.4 10.5

Influence of frailty-related diagnoses, high-risk prescribing in elderly adults, and primary care use on readmissions in fewer than 30 days for veterans aged 65 and older.

To determine the effect of two variables not previously studied in the readmissions literature (frailty-related diagnoses and high-risk medications in...
98KB Sizes 0 Downloads 0 Views