Incidence and Cost of CAP in a Large Working-Age Population
Published Online: July 19, 2012
Machaon M. Bonafede, PhD, MPH; Jose A. Suaya, MD, PhD, MPH; Kathleen L. Wilson, MPH; David M. Mannino, MD; and Daniel Polsky, PhD
Patient demographic characteristics were measured on the index date and included age, gender, US Bureau of Census region of residence, health plan type, insurance plan capitation, population density status (urban vs rural), industry type, employee classification (hourly versus salary), and union classification (union vs non-union). Clinical variables calculated during the baseline period included the Deyo Charlson Comorbidity Index (CCI); comorbidities of chronic obstructive pulmonary disease (COPD), congestive heart failure (CHF), cancer, cardiovascular disease (CVD), other musculoskeletal surgery, diabetes, and asthma; pre-period use of antibiotics, long-acting beta-agonists, short-acting beta-agonists, and systemic corticosteroids; indicator of whether there was a hospitalization in the 1 month prior to index date; 6-month pre-period costs for inpatient, outpatient, and pharmacy; and time observed for follow-up.
Incidence of CAP
The incidence of CAP for the 5-year study period was calculated based on all the eligible benefi ciaries in the Commercial Database between January 1, 2003, and December 31, 2007, that had at least 6 months of uninterrupted enrollment preceding the index date for pneumonia or controls. The numerator was the number of cases during the study period. The denominator was the total enrollment time during the study period in the Commercial Database among potentially eligible patients. Incidence rates were reported as CAP cases per 1000 person-years and stratifi ed by gender and age categories.
Excess Cost of CAP
In order to estimate the direct medical cost and productivity burden of pneumonia, we generated a control group from all beneficiaries without a pneumonia diagnosis using nearest neighbor propensity score matching without replacement in a 3:1 control:case matching ratio. The propensity score was estimated by using a logistic regression modeling the presence of pneumonia as the dependent variable with the following covariates: index year; quarter; age; gender; geographic region; plan capitation status; industry; length of follow-up; baseline Deyo CCI; and baseline comorbid conditions, including asthma, cancer, CVD, CHF, COPD, diabetes, and HIV. Propensity score matching was performed and reported separately among the subset of patients (and controls) with work absence and short-term disability data. Despite the propensity score match, imbalances in risk will remain, and with the highly skewed distribution of costs, these imbalances may have an important infl uence on the estimate of excess costs. Therefore, in addition to the excess costs estimated from the observed differences between the pneumonia and non-pneumonia matched groups, we estimated excess costs in multivariate analyses by ordinary least squares (OLS) and generalized linear model (GLM). Separate models were run for total inpatient cost, ambulatory care excluding pharmacy cost, and pharmacy cost due to the distinct cost distribution for each of these categories. These predicted costs were then summed in order to assess total costs. Similarly, we used separate models to analyze absenteeism and short-term disability. Two-part models were used for inpatient costs, absenteeism, and short-term disability.
For GLM, the log link and gamma variance functions were used. The control variables for all multivariate models included: age; gender; insurance type; region; urbanicity; employee class (union vs non-union); employment status; employee relationship; Deyo CCI score in pre-period; COPD; CHF; cancer; CVD; other musculoskeletal surgery; diabetes; asthma; pre-period use of antibiotics, long-acting beta-agonists, short-acting beta-agonists, systemic steroids; indicator of whether there was a hospitalization in the 1 month prior to index date; total inpatient costs in pre-period; total outpatient costs in pre-period; total pharmacy costs in pre-period; and time observed for follow-up. All cost estimates are in 2008 US dollars.
National Projections of CAP
To quantify the total annual cost of CAP that occurs in the US population aged 18 to 64 years, we multiplied our estimate of cost of an episode of CAP by our estimate of the incidence of CAP and by the number of people in the United States in this age group. We quantify total annual costs of CAP with and without productivity costs. We express our range of methodological uncertainly by producing estimates for each of our 3 methods of estimating medical and productivity costs. We quantify uncertainty to stochastic variation in our estimates through Monte Carlo simulation based on the means and standard errors of inputs with distributional assumptions. Because each input could be skewed to the right and had to be non-negative, we fi tted lognormal distributions to each major input and ran a simulation with 20,000 iterations. The simulation was done in Crystal Ball, Fusion Edition, version 11.1, and Excel 2007 software packages.9,10
Table 1 contains a summary of the incidence of CAP among patients in the MarketScan Commercial Database, in total and stratified by age. Annualized overall incidence for this population was 4.89 cases per 1000 person-years. This incidence calculation was based on the 402,831 cases of CAP identified from 2003 to 2007 in our sample of 36.25 million commercially insured adults aged 18 to 64 years, with a total enrollment time of about 83 million years of risk. We project this annualized rate to the current US adult non-elderly population to arrive at an estimate of 947,000 cases of CAP in 2010 among adults aged 18 to 64 years. This analysis suggests increasing incidence with age; incidence among patients aged 55 to 64 years was 3.3 times the incidence of the group aged 18 to 34 years.
Demographics and Baseline Clinical Characteristics of CAP Sample
PDF is available on the last page.