Overview of methods used to estimate US health spending by age, sex, payer, type of care, and 165 causes of illness: 1996 – 2013

Although estimates of total health expenditure for the United States are available, comprehensive data on the amount spent on different diseases and by which populations are still lacking. To fill this gap, we have embarked upon a three year project to estimate annual health expenditure in the US disaggregated by disease or injury as classified by the Global Burden of Disease (GBD) 2013 as well as by payer, age, sex and type of healthcare, from 1996 to 2013.

The first step of our analytic process involves attaining total US health expenditure by type of service. We use estimates from the National Health Expenditure Accounts (NHEA). We parse the total expenditure into inpatient services; office-based outpatient services; hospital outpatient services; dental services; emergency department services; long-term nursing and home care services; prescribed pharmaceuticals; over-the-counter pharmaceuticals; medical devices; public health and prevention services; and health system administration including insurance administration. We use encounter-based microdata to estimate expenditure and utilization for each type of service. We translate International Classification of Disease (ICD) coded data to GBD diseases, injuries and conditions. We smooth estimates over age and time stratified by healthcare type, disease and sex using a two-dimensional hierarchical Bayesian regression-based process to produce plausible time-trends. Sources with known biases are adjusted to be nationally representative of all payers and to include out-of-sample specialty mental health facilities. Sources which rely on charge data to estimate expenditure are adjusted to reflect payments using function-disease-payer-specific adjustment ratios. The data is then adjusted for the presence of comorbidities that may enlarge the expenditure due to the treating of multiple diseases within a single health system encounter. Estimates are compared to the NHEA totals to create function-specific scaling factors. We use smoothed estimates of total encounters and the final estimates of expenditure to calculate average price per encounter. Finally, we bootstrap the microdata to produce confidence intervals propagated throughout the process for every estimate. For each of the eight healthcare types, we produce expenditure, utilization and price estimates for 165 GBD conditions, three payers, and 38 age-sex groups for each year between 1996 and 2013.

The novel methods discussed here meld ten distinct sources of data, including data from insurance claims, government budgets, and health facility and household surveys to estimate health spending by age, sex and disease.  Ultimately, these estimates can contribute to better health resource allocation policies and allows for more in depth analysis of the drivers of expenditure for specific diseases and populations.