Intraclass Correlation and Variance in the Characteristics of Primary Care Patients Managing Chronic Medical and Behavioral Conditions.

Background To avoid statistical errors, researchers who recruit patients from selected medical practices and analyze them at the individual level need to account for the clustered nature of their sample. This is most often done using the intraclass correlation coefficients (ICCs), a measure of how strongly subjects recruited from the same cluster (in this case patients from a clinic) resemble each other. Aims The aim is to support the design of cluster-randomized studies by supplying estimates of variance and ICC of various measures using a population of patients from multiple primary care clinics. Materials and methods ICCs were extracted from a large cluster-randomized pragmatic clinical trial of adult primary care patients managing multiple chronic conditions, the Integrating Behavioral Health and Primary Care study (IBH-PC). IBH-PC collected demographics and patient-reported health outcomes on over 3,000 adults from 44 primary care practices in 13 states across the US. We present estimates of the standard deviation and ICC for gender, race, ethnicity, marital status, employment, income, education, social determinants of health, PROMIS-29 functional status, Duke Activity Status Index (DASI), nine-item Patient Health Questionnaire (PHQ-9) depression score, Generalized Anxiety Disorder (GAD-7) anxiety score, Asthma Symptom Utility Index, restricted activity days, medication adherence, health care visits in the past month, emergency room visits in the past year, hospital days in the past year, perception of quality and patient-centeredness of care, alcoholic drinks per month, and the GAIN substance use disorder screener. Results ICCs varied broadly with the highest values found for race and income and the lowest for short-term estimates of the GAIN. Conclusions These values can be used to inform the design, especially power estimates and sample size requirements, of future studies.