Study Design, Population, Settings
This observational cohort study was conducted in the member population of Kaiser Permanente Southern California (KPSC), an integrated healthcare organization with more than 7500 physicians, 15 medical centers, and 231 medical offices. The KPSC system provides comprehensive healthcare to more than 4.6 million racially and socioeconomically diverse members residing within 7 counties of Southern California. KPSC coordinates care through regionwide electronic medical records (EMRs) that capture information on outpatient visits and inpatient stays received at KPSC-owned and KPSC-contracting facilities. The research database also includes administrative claims for members who receive out-of-network clinical care. All other patient data including demographics, lifestyle factors (e.g., smoking, body mass index [BMI]), socioeconomic factors, and comorbidities are stored in the EMR. The study was approved by the KPSC Institutional Review Board.
Inclusion and Exclusion Criteria
This study included all KPSC members who received a SARS-CoV-2 positive nucleic acid amplification test between March 1 and August 31, 2020 (N=67,354). Patients were excluded from the study if they were not KPSC members (N=3338) or had an unknown gender (N=5).
Outcomes, Exposure, and Measures
The primary outcome was overall healthcare utilization in the 180 days (in 30-day time intervals) following a test-confirmed COVID-19 diagnosis. All encounters occurring on the same day of the COVID-19 test were included. Patients were followed from the time of the COVID-19 test to 180 days after, death, or end of membership, whichever came first. To account for varying follow-up periods among patients, utilization was measured as the number of visits per 30 person-days.
Since most of the healthcare utilization occurred in the first 30 days following the COVID-19 diagnosis, we performed more detailed analyses on the visits during this first time window. Specifically, we evaluated the types of visits, such as telehealth (phone or videoconference appointments), in-person office visit, assisted care (e.g., hospice/skilled nursing facility), ED, urgent care, or hospitalization. For telehealth and in-person office visits, we further assessed the visit specialty (primary care or specialty care) and type of provider (physician, physician assistant/nurse practitioner, nurse, or mental health counselor).
In addition, we assessed the reasons for visits using the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) codes associated with the visits in the first 30 days. ICD-10-CM codes were aggregated into clinically relevant body system categories (e.g., respiratory, circulatory, endocrine systems) using the Healthcare Cost and Utilization Project (HCUP) Clinical Classifications Software Refined (CCSR) for ICD-10-CM-coded diagnoses.14COVID-related diagnosis codes25 (Supplemental Table 1) were grouped into a separate category.
We calculated the average rates (per 30 person-days) and 95% confidence intervals (CI) for the total number of visits in 30-day intervals up to 180 days after COVID-19 diagnosis. For utilization in the first 30 days, we also calculated rates for the types of visits, visit specialty, type of provider, and reasons for visits. Rates for visit specialty and type of provider were only calculated for telehealth and in-person office visits.
For the total number and types of visits in the first 30 days, we calculated rates within subgroups defined by age category (<18, 18–39, 40–64, ≥65), gender, race/ethnicity (Asian, black, Hispanic, white, other), and BMI category (underweight <18.5, normal 18.5–24.9, overweight 25–29.9, obese 30–39.9, severely obese ≥40).
Poisson regression was used to identify factors associated with higher utilization in the first 30 days. We estimated rate ratios (RR) and 95% CIs for the association between overall utilization and age category, gender, race/ethnicity, BMI category, smoking status (current, former, never), income (<$40,000, $40,000–79,999, ≥$80,000), college education, and Elixhauser comorbidity index (0, 1, ≥2). We included follow-up time as an offset in the model to account for varying observation times (due to censoring from death or end of membership during a given window).
Rate calculations and Poisson models were performed separately for patients who were and were not hospitalized within 30 days from the COVID-19 diagnosis. All analyses were conducted using SAS 9.4.