below. Fourth, the variable representing total indemnity payments on a claim does not bundle together payments made by defendants’ insurers and payments made by state-run patient compensation funds (PCFs) for the same incident. Nine states have patient compensation funds, which function as a secondary layer of liability insurance, paying judgments and settlements above a certain amount. The PUF lists PCF payments separately, but does not provide a means of directly linking them to the primary payment in the same case via a case identifier. We used a probabilistic matching method employed by Studdert and colleagues to link PCF payments to insurer payments relating to the same incident, but this algorithm may have linked payments imperfectly.13 2.1.2. Definition of “Medicare” Group Because we could not precisely identify Medicare beneficiaries within the PUF, we constructed two different groups of elderly claimants. The main analysis compares patients 70 years and older at the time of the incident underlying the claim to patients under age 60 at the time of the incident. (Claimants aged 60-69 at the time of the incident are omitted from the analysis because that group 14 contains both Medicare enrollees and non-enrollees.) A sensitivity analysis compares patients 60 years and older to those under 60. It should be noted that End-Stage Renal Disease patients are likely misclassified in our analysis. Though Medicare eligible, they are likely to be disproportionately under the age cutoffs that define the “Medicare” groups in our analysis. Thus, our analysis treats them as non-Medicare patients unless they exceed the age thresholds. Persons who qualify for Medicare as Disabled but are not elderly are also erroneously classified as non-Medicare beneficiaries in our analysis. ESRD and Disabled patients are not identifiable in the NPDB, making it impossible to correct any potential misclassification. Because ESRD patients represent less than 1% of Medicare beneficiaries14 and about 45% of those patients are over age 65,15 the number of persons misclassified—and thus the effect of the misclassification on our results—is small. In contrast, non-elderly Disabled persons comprise more than 16% of Medicare beneficiaries. We have no way of determining what proportion of NPDB reports they represent, and thus what the magnitude of the potential misclassification bias in our analysis might be. To account for these issues, throughout the report we have referred to our Medicare group as representing “elderly” Medicare enrollees, rather than all enrollees. 2.1.3. Analytical Methods We used ordinary least squares (OLS) regression, chi-square tests, and t-tests to examine the relationships between key outcome variables and Medicare enrollee status. In regression models predicting indemnity payments, payments were logged to reduce nonnormality. To examine time trends, we modeled year as a continuous variable (coded 1 through 12) and used OLS regression. We also ran Cuzick’s nonparametric test for trend across ordered groups on the 12-point time variable. This test, which is implemented in Stata using the nptrend command, is an extension of the Wilcoxon rank-sum test. It tests the hypothesis that the values of a variable with a natural ordering systematically increase or decrease over levels of another variable. All analyses were run in Stata version 13.1. We included state PCF payments in calculations of indemnity payment levels, but excluded them from all other analyses, such as frequency counts, to avoid double counting of cases relating to the same incident. Our analysis shows the distribution of claims by age group, but does not independently control for differences in health status or healthcare utilization. Because of their greater utilization of medical care, the elderly will, on average, have greater exposure to malpractice (though that may not hold true in the realm of missed and delayed diagnoses). We do not test for whether age in itself makes a person more or less likely to appear in the NPDB (for example, because older persons are more or less likely to sue or prevail in lawsuits than younger persons). 15 2.2. Results 2.2.1. All Claims 2.2.1.1. Sample Characteristics Our analytical dataset contained 116,965 paid claims reported between January 1, 2005 and December 31, 2015 (Exhibit 4). About 94.2% of these (110,199 claims) were payments made by insurers, and the remaining 6,766 payments (5.8%) were made by state PCFs. Among the non-PCF payments, 92.2% were paid in the name of a non-trainee MD, 7.1% were paid on behalf of a non-trainee DO, and less than 1% were paid on behalf of a trainee. Nearly all (95.7%) were made pursuant to a settlement. Nearly a third (32.2%) of claims involved deaths and almost another third (30.6%) involved serious, permanent injuries. The majority (56.6%) of incidents occurred in the inpatient setting. Across all payments, the median total payment was $211,046 in 2015 dollars. Twenty-five percent of payments were $73,834 or less and 75% were $500,321 or less. The 99th percentile of payments was $2.75 million. The most common clinical classifications of the incidents underlying paid claims were diagnostic related (31.5%), surgery related (27.0%), and treatment related (19.7%). Among diagnostic-related claims, 53.5% concerned an alleged failure to diagnose, 26.6% a