Association of Past and Future Paid Medical Malpractice Claims

Key Points Question Do prior paid medical malpractice claims predict future paid claims? Findings In this retrospective, case-control study including all 881 876 physicians licensed to practice in the US at the time of the study, physicians with 1 paid claim (regardless of specialty) were almost 4 times more likely to have 1 or more paid claims in the next 5 years compared with physicians with no prior paid claims. The likelihood of future claims rose monotonically with the number of prior claims and was unaffected by whether paid claims were publicly disclosed. Meaning The findings of this study suggest that paid medical malpractice claims are not random events; timely noncoercive intervention has the potential to reduce future claims.

For low-skill physicians, v will be large, most claims will be true positives, and PPV will be high. For high-skill physicians, v will be low, PPV will be lower and more claims will be false positives, with no underlying negligence. This can be seen as a simple application of Bayes' theorem: the lower the prior probability of negligence, the lower the posterior probability that an observed paid claim reflects actual negligence.
While the model focuses on paid claims, this tendency will be stronger for all claims, many of which will be closed without payment, than for paid claims. If most physicians are high-skill, this could help explain the common physician belief that most med mal claims are false positives. This belief could be true for most physicians, yet false for most paid claims. While we have data only for paid claims, it is possible that the fraction of claims brought which close without payment is higher for high-skill physicians, who may rely on this experience to conclude that claims against other physicians lack a strong basis. A similar dynamic may affect high-skill physicians who are involved in a medical malpractice case where there are also claims against low-skill physicians. A high-skill physician, who is exonerated or dropped from the case may conclude the case lacked merit, even if there was a subsequent payment on behalf of the low-skill physician. 1 Moreover, high skill physicians may cluster in particular hospitals and geographic areas 2 Thus, high-skill physicians spend most of their time around other high-skill physicians, which could reinforce their experience-based belief that most med mal claims reflect bad luck, rather than bad care.
Note that some low-skill physicians will be "lucky" and not experience claims in the prior period, but will still be more likely to experience future period claims, while some high-skilled physicians will have been unlucky in the prior period and faced a paid claim, but will remain unlikely to face a claim in the future period. Thus, this study of future risk faced by apparently skilled versus less-skilled physicians (both defined based on prior period claims) will understate the true effect of physician quality on med mal risk.
The model can be extended by adding a time dimension. In each year, each physician either will or will not experience a paid claim. If the physician does not experience a paid claim, the best estimate of the likelihood of negligent care r(q) will go down; if the physician experiences a paid claim, this estimate will go up. The more prior paid claims, and the more recent they are, the higher the PPV will be for the most recent paid claim.
© 2023 Hyman DA et al. JAMA Health Forum.

Simulation Methodology
The primary simulation methodology uses a 5-year period to measure past claims and a 5-year period to measure future claims, with 2009-2013 as the prior period and 2014-2018 as the future period.
Note that we compute average claim risk across a number of years, during a time period when the number of claims was dropping steadily (see eFigure 1). Allowing claim risk to vary by year would produce slightly different counts. The approach is similar for other prior and future periods.
The average annual state-specific risk of a paid medical malpractice claim is defined as:

fs = claimss/docss
Here claimss is the average number of claims in state s over 2009-2018, and docss is the average number of physicians in state s over 2014-2016 (first 3 years of the "future" period).
For each year in the future period, the study takes docss draws of claims from a binomial distribution, with probability fs for each draw. This produces a number futurest of paid claims for each state s and future year t, where futurest is a random variable with expected mean = claimss, but the actual values will vary across simulations and future years. For each future state*year, the study randomly assigns each of these claims to one of the physicians in that state. The study then counts the number of suits for each physician for each future year, and sums these physician-specific counts over the future period. Since fs is small (the national average is around 1%), most physician*years will have zero claims, a few (approximately fs) will have one claim, a smaller number (approximately fs 2 ) will have two claims, and so on.
If claims arrived at random, the likelihood that a physician would receive one claim in 5 future years will be about 5*fs (about 5% for an average state), the likelihood of two claims will be about actual values from Table 1, simulation row, which shows a 4.78% likelihood that a physician will receive 1+ future claims in the next 5 years and an 0.12% likelihood of 2+ future claims.) We run each simulation 10,000 times, and thus obtain a mean number of physicians who will receive 1 paid claim (say) in the future period, and a distribution around that mean. We use this distribution to determine the 95% confidence interval around the mean number.
We also use NPDB data to determine the actual number of physicians who receive 1, 2, 3, etc.
paid claims during the prior period and the future period.

eTable 1. Risk of a Paid Claim by State
Table shows the risk that physicians in a particular state will have a paid med mal claim, state-specific risk relative to national average, overall risk rank, and whether the state has public disclosure of paid claims. Paid claims are annual average by state over 2011-2015, rounded to nearest whole number, using the licnstat variable in NPDB. Active physician counts are based on data from AHRF, averaged over the same time period. States are sorted by risk level, from highest to lowest.
(1) Number of licensed Illinois physicians in indicated specialties and surgical sub-specialties, and paid claim risk over 1990-2016, relative to the average risk for all physicians with known specialty. Relative to the AHRF specialties shown in eTable 2, the Illinois data does not include plastic surgery (included in general surgery in Illinois); gastroenterology, cardiology, and pulmonary medicine (all included in internal medicine in Illinois), and radiation and oncology (included in radiology in Illinois).  Table 1 is that in this table we report separately data for physicians with 3 versus 4+ paid claims during the baseline and future periods. Panel B shows the predicted average probability across all states for a physician to have the indicated numbers of paid claims during the future period, if claims arrive at the average state-specific rates observed for the baseline period, but otherwise at random (independent of physician skill) and all physicians active during the baseline period remain active in the future period. Probabilities are mean values from 10,000 simulations, averaged across states, with states weighted by average number of active physicians in the baseline period. Panel C shows the predicted numbers of physicians with the indicated numbers of paid claims during the future period under the same assumptions as Panel B. Predicted numbers are mean values from 10,000 simulations. Panel D shows ratio of actual to predicted claims during the future period.   Table parallels text Table 2, and shows the ratio of the risk for distribution of 1+ (Panel A), 2+ (Panel B) and 3+ (Panel C) paid claims over varying future periods among active practicing physicians based on each physician's number of paid claims over the baseline period), compared to the likelihood assuming claims arrive randomly at the average rate during the baseline period. Base probabilities with random claim arrival assume that all future period claims involve physicians active during the baseline period, and are mean values from 1,000 simulations, assuming claims arrive randomly at state-specific rates. Ratio for physicians with 0 baseline claims for having 1+ future claims is less than 1.00 due to lower risk for these physicians plus secular decline in paid claim rate, but we normalize this ratio to 1.00 for each specified future period to simplify interpretation.   -18 2016-18 2015-18 2014-18 2013-18 2012-18 2011-18 2010-18 baseline period 2013-17 2012-16 2011-15 2010-14 2009-13 2008-12 2007-11 2006-10 2005-