We followed 44,106 individuals who were given the choice between a low- or high-deductible Preferred Provider Organization (PPO) plan through a large multinational company between the years of 2001 and 2006. Annual deductibles for these plans ranged from U.S.$250 to U.S.$2,500. The high-deductible plans were provided to employees at no cost to them; employees who opted for low-deductible plans had to pay a share of the premium. A comparison group of 60,107 individuals enrolled in a low-deductible PPO (U.S.$200) plan served as a control.
Each person in our sample had no prior history of chronic disease, was under 65 years old, enrolled from 2 to 6 years of the observation period, and eligible for one or more selected preventive screens according to criteria from the National Committee for Quality Assurance Healthcare Effectiveness Data and Information Set (HEDIS). These included endoscopy (including both flexible sigmoidoscopy and colonoscopy; males and females 50 and older), fecal occult blood tests (FOBTs) (males and females 50 and older), lipid screening (males 35 and older, females 45 and older), Papanicolaou (pap) smear screening (females 18–64), and mammography (females 40–64).
We chose these five screens because they are widely regarded as effective and because they can be identified based on International Classification of Diseases, Ninth Revision, Clinical Modification, Current Procedural Terminology, and Health Care Financing Agency Common Procedure Coding System codes, which are available in health insurance claims data.
In some cases, the HEDIS guidelines indicate that a screen should be administered less frequently than once a year. In those cases, utilization rates in our study, which are based on annual use, are lower than would be the case if we used multiyear utilization rates. Because we calculated utilization rates consistently across plan types and over time, this discrepancy does not bias our results.
Literature provided to us by the employer states that beginning January 1, 2003, the plan would offer zero-deductible coverage (or “FDC”) of “preventive medical tests and routine physical examinations.” This reduced the cost of preventive services to their postdeductible levels; coinsurance rates ranged from 20 to 30 percent of total costs for the affected PPOs, varying by plan, year, and network. The plan literature directly cites pap smears and mammograms. Where coverage of preventive services was not explicitly stated in the plan descriptions, we determined coverage empirically by comparing PPO enrollees' mean out-of-pocket expenses before and after the introduction of FDC ().
Out-of-Pocket Expenditures on Preventive Screens
In this study, we exploited three types of variation—changes in coverage of preventive services over time (within a plan); coverage of different preventive services, that is, endoscopy versus other preventive services; and differences among individuals' enrollment choices—to test two hypotheses:
- First, we hypothesized that FDC would be associated with increased utilization of covered preventive services, because FDC lowers the effective price to consumers.
- Second, we hypothesized that after controlling for demographics and secular trends, the expanded coverage would be more effective in increasing utilization among healthy low-deductible enrollees than high-deductible enrollees, because the value of preventive care to high-deductible enrollees remains lower than actual or perceived dollar costs.
To test the first hypothesis, we used multivariate logistic regression (clustering error by individuals) to calculate the adjusted utilization rates with and without FDC for the five screens. The dependent variables were (1) lipid screens, (2) mammograms, (3) pap smears, (4) FOBTs, and (5) endoscopy (not subject to FDC). Each dependent variable was coded as 1 if the service was used in a particular year, 0 otherwise, a dummy variable for FDC (0 before 2003; 1 otherwise). As control variables we included the enrollment year, gender, and year of birth. We were not able to control for other important variables, such as socioeconomic status, education, and occupation, because they were not in our dataset. To test the second hypothesis, we repeated the regressions separately for low- and high-deductible enrollees.