Data sources and study population
We used the IBM® MarketScan® Commercial Claims and Medicare Supplemental Databases from 2011 to 2019 to describe characteristics of children and adults enrolled in non-capitated health plans who were tested for RSV or had a medical encounter with an RSV diagnosis. MarketScan provides administrative claims data for employees and their dependents who are covered by employer-sponsored private health insurance in the US. Enrollees in capitated health plans were excluded because providers are reimbursed with lump sum payments rather then for individual services, which may compromise complete capture of medical encounters via claims and thus result in lower data quality). The MarketScan database provides patient-level longitudinal data for a national sample of more than 100 million enrollees in private insurance plans, including detail about their demographic characteristics, medical encounters, and dispensed prescription drugs. As with other national insurance claims databases, MarketScan has been used to derive population-based RSV incidence estimates [21,22,23,24].
For the same time period, we analyzed NREVSS to assess the concordance in test type distribution and test positivity with MarketScan. NREVSS is a voluntary surveillance system that collects aggregated weekly lab data from participating U.S. laboratories to monitor respiratory virus circulation [17, 25]. NREVSS provides quantity and results of diagnostic tests for RSV by test type (i.e., PCR, viral culture and antigen -a combination of rapid antigen and immunofluorescence) and by 10 regions classified by the U.S. Department of Health and Human Services (HHS) [25].
Study design and inclusion criteria
For the assessment of RSV test distributions, we enrolled all children and adults in MarketScan who had one or more RSV tests from out- or inpatient settings. We identified RSV tests based on claims using Current Procedural Terminology (CPT) codes (Additional file 1: Table S1). Because some CPT codes for RSV testing are not specific [e.g., 87,798 Infectious agent detection by nucleic acid (DNA or RNA), not otherwise specified], we evaluated 3 alternative definitions to identify RSV tests, including a broad (i.e., all potential CPT codes), strict (i.e., specific CPT codes for RSV testing and nonspecific CPT codes accompanied by relevant diagnoses (e.g., a respiratory illness) on test claims (Additional file 1: Table S2)) and a very strict definition (i.e. only specific CPT codes specifying a test for RSV).
Frequency of RSV tests
We estimated the frequencies of RSV tests in both MarketScan and NREVSS. In MarketScan, if multiple tests were conducted for the same patient on the same day (< 1%), the test with the highest sensitivity was retained (PCR > viral culture > antigen). Though an antibody test is not recommended to diagnose RSV infections, we report the proportion of antibody tests to fully capture testing practice. To be eligible for this category, we required patients to not have other RSV tests on the same day. NREVSS does not identify tests at the level of patients, thus only frequencies of tests are reported.
Positivity of RSV tests
Test positivity was obtained directly from test results available in NREVSS. In MarketScan, we approximated the proportion of RSV positive tests based on diagnoses codes on laboratory claims or other medical encounters adjacent to the laboratory claims date, because the actual test result is not available from claims data. To do so, we searched for outpatient encounters with primary or secondary diagnosis codes indicating an RSV infection within ± 7 days of the RSV test claim or for inpatient encounters that overlapped with or followed an RSV test within 3 days. These time windows were chosen to accommodate turn-around times of RSV tests, which depend on test types and capacity for testing [20, 26, 27]. Patients were required to have continuous insurance coverage during these time windows to fully capture medical encounters.
We compared the test distributions and proportion of positive tests across strata estimated from claims data with those reported by NREVSS. The stratified analyses considered calendar year, quarter, HHS region and test type (all available in both databases) and age, assumed test indication and clinical setting in MarketScan. We obtained test indications from diagnoses codes on test claims and group diagnoses based on disease severity: we gave priority to infections with severe complications (i.e., septicemia/sepsis, respiratory distress/failure), followed by LRTIs, asthma exacerbations, upper respiratory tract infections (URTIs) including otitis, respiratory symptoms, and others (Additional file 1: Table S2).
For comparison of secular trends in testing, we standardized annual data (2011–2018) to the number of enrollee-years in 2019, because the MarketScan population changed over time due to different health plans providing data in each year. The size of the source population for the RSV tests in NREVSS is unknown.
Influence of positivity variation on RSV incidence estimates
RSV test positivity rates are commonly used to estimate overall RSV incidence by imputing the presence of RSV among respiratory tract infections that carry no code for the responsible pathogen. To illustrate the impact of differences in the population from which test positivity rates are obtained and the population to which they are applied to estimate RSV incidence, we calculated the corrected RSV (cRSV) incidence. The total number of cRSV encounters was the sum of all acute respiratory infections with coded RSV (ARIRSV), which was directly obtained from MarketScan, plus the unknown number of ARIs without coded pathogen where the pathogen was likely RSV (\({\overline{ARI} }_{RSV}\), Eq. 1).
$$\mathrm{cRSV \, incidence}=\frac{ {ARI}_{RSV} + {\overline{ARI} }_{RSV}}{\#\mathrm{ enrollee \, months}}$$
(1)
To obtain \({\overline{ARI} }_{RSV}\), we assumed that all ARIs without coded pathogen in MarketScan are a combination of ARIs with negative RSV test plus untested ARIs. The number of ARIs with negative RSV test was directly estimated from the RSV test positivity (e.g., if test positivity is 10% and we count 100 ARIRSV, the number of tested ARIs that were negative is 100/10%-100 = 900). Subtracting this estimated number of RSV-negative ARIs from all uncoded ARIs provided the number of untested ARIs. This number of untested ARIs was then multiplied with varying-test positivity rates (assuming 10%, 20%, 40%, 60%, 80%, or 100% of the positivity of observed RSV-tested ARIs) to obtain \({\overline{ARI} }_{RSV}\). This range reflects our assumption that positivity among untested ARIs is lower than that of tested ARIs because clinicians likely order a test for patients where RSV is suspected. An example to illustrate this calculation is presented in the Additional file 1: Supplementary Methods.
We compared the percent misclassification of corrected incidence rates across strata when using MarketScan stratum-specific (age, setting and indication) positivity rates versus the NREVSS average positivity rate (Eq. 2).
$$\mathrm{\%misclassification}=\frac{ cRSV \,incidence \,using \,NREVSS-cRSV \,incidece \,using \,MarketScan}{cRSV \,incidece \,using \,MarketScan}*100\%$$
(2)
All analyses were conducted for either antigen or PCR tests in each stratum. All analyses were conducted with SAS Studio 3.8 (SAS, Cary, NC).