Study Design and Data
We conducted a retrospective matched cohort study using data from the 2011–2015 Longitudinal Medical Expenditure Panel Survey (MEPS). The MEPS is a well-established source of national data on access to health services, health care utilization, and health care expenditures.18 Medical conditions of MEPS respondents are collected in a narrative form and then coded by trained coders using the International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) codes. The Clinical Classification Category (CCC) codes—clinically meaningful combinations of ICD-9-CM codes classified into 274 mutually exclusive categories created by AHRQ—are also provided in the MEPS data.19 We used 5 years of overlapping MEPS Longitudinal Survey data (2011–2015; panels 16–19). For each panel, we merged data from Medical Conditions19 and Hospital Inpatient Stay Files20 using a unique person identifier. This study used deidentified and publicly available data and was deemed exempt from review by the University of Florida institutional review board.
Study Cohort: Ambulatory Care Sensitive Conditions
Our initial analytic sample included 24,296 US adults aged 18 years or older with any of the following 5 ACSCs: (1) hypertension, (2) type 2 diabetes, (3) heart disease, (4) asthma, and (5) chronic obstructive pulmonary disease (COPD).21 We used a two-step approach to identify participants with ACSCs. First, we identified these ACSCs using patient self-reported data on medical conditions. Participants who responded “Yes” to one or more of the following question “Have you ever been told by a doctor or other health professional that you had the health condition?” for hypertension, type 2 diabetes, heart disease (coronary heart disease, angina, myocardial infarction, and other unspecified heart disease), asthma, or COPD (emphysema and chronic bronchitis) were identified. Next, we validated ACSC among these participants using specific ICD-9-CM or CCC. A full list of these codes appears in Online Appendix Table 1. Having identified individuals with ACSCs, we further excluded patients if (1) they reported never receiving instructions about a specific illness or health condition (n = 2087) because receiving instruction from a health care provider is a prerequisite to having teach-back experience, (2) the patient was a pregnant female at the time of survey (n = 2273) due to the likelihood of hospital admission for delivery, (3) the patient was diagnosed with any cancer (n = 2185), because they may have unusual patterns of health services use due to severe conditions, or (4) the health services utilization data were missing (n = 3641).
Main Exposure: Patient-Reported Teach-Back Experience
After identifying the initial ACSC cohort, we determined their exposure to teach-back. For this purpose, we used a question from the Consumer Assessment of Healthcare Providers and Systems (CAHPS®), which measures the patient’s perceived quality of health care services received in the last 12 months.22 Teach-back experience was assessed and categorized according to patients’ response to the CAHPS® question asking, “How often doctors or other health providers asked you to describe how you are going to follow their instructions?” Individuals were defined as having the teach-back experience if they responded that they “always” received a teach-back. The question is administered once and it represents a summative assessment of all visits in the past 12 months. This definition of teach-back exposure is similar to that recommended by the AHRQ and used in previous studies to track teach-back 8,23. ACSC patients who responded to the question otherwise were considered not exposed, making the conservative assumption that a patient derived no benefit from the teach-back experience if they were not exposed to the teach-back technique. After identifying patients with teach-back experience, our total cohort included 14,110 patients with ACSCs.
Outcomes: Hospital Admission
Our primary outcome of interest was hospital admission during year 2 of follow-up (panel rounds 3–5). We identified and classified hospital admissions into two types: (1) “all-cause” which included hospital admission for any reason and (2) “ACSC-related” which included hospital admission with primary cause of visit listed as an ACSC condition. Secondary outcomes included (1) a second admission within 12 months of discharge of the first hospital admission and (2) length of stay (i.e., time from hospitalization until discharge) to assess the severity of exacerbation. Given the acuity of ACSCs3,4,5 and more than 95% of those hospitalized were discharged within 30 days, we censored those having more than 30 days of inpatient stay.
Demographic and health characteristics that may affect health services utilization and confound the relationship between teach-back experience and hospital admission were included as covariates.24,25 These included age, sex, race/ethnicity, education, family income level, employment, marital status, census region, health insurance type, having a usual source of care, general health status, the SF-12 Physical and Mental Summary scores, current smoking status, and obesity (defined as a body mass index greater than 29.9 kg/m2).
Each patient with an ACSC and with teach-back experience was statistically matched with a similar patient who did not have the teach-back experience. Participant matching was performed using the nearest-neighbor propensity score matching technique to match teach-back individuals to control individuals.26 Propensity scores were computed using baseline individual characteristics listed above and the panel numbers. The variables in the propensity score model were selected using a stepwise procedure to predict the probability of having the teach-back experience.
Baseline cohort characteristics were described using proportions and means as appropriate. We tested for statistically significant differences in demographics and health status by teach-back experience using a chi-square or Fisher’s exact test for categorical variables and t tests for continuous variables. We used Poisson regression models to test whether teach-back experience was associated with hospital admission (for all-cause as well as ACSC-related condition). The association between teach-back experience and duration of the first hospital admission was estimated using Kaplan-Meier survival curves, and the statistical significance of differences by teach-back experience was determined using the log rank test.27 The association between teach-back experience and risk of second admission within 12 months of initial hospital discharge was estimated using Cox proportional hazards models.
We also tested the sensitivity of results to the operational definition of teach-back experience. First, patients were defined as being exposed to the teach-back experience if they reported “usually” and “sometimes” having a teach-back (rather than just “always”, as in the primary analyses). Sensitivity analysis was also conducted to evaluate the effect of English proficiency and for this purpose we excluded those who were not born in the USA and having difficulties in communication using English. We also performed subgroup analyses to examine the differences in the outcome measures by the five specific ACSCs and combination of more than one ACSC; individuals were matched according to the specified ACSCs separately. We used an alpha level of 0.05. The PROC SURVEY in SAS® (9.4, SAS Institute) and Complex Survey procedures in SPSS (24, IBM) were used for analyses accounting for the MEPS sampling weights and the complex survey design.