No rule is free of exceptions. Although quality criteria define which patients should receive which care processes, some patients will not receive a particular item of care for any of a number of reasons: for example, patients with acute myocardial infarction should, by standard, receive aspirin on admission, but this is of course foregone if the patient is allergic to aspirin. In this case, "high quality" can be defined as care that appropriately distinguishes exceptions from regular cases and tailors treatment to the individual patient. Unfairly penalizing such cases as "failures" to provide standard care should be avoided.
Three methods of handling exceptions to quality indicators can be considered. First, these cases can be excluded from the denominator in calculating the adherence rate, thereby providing a "pure" sample of standard patients. In our study also, denominators for some indicators were made narrower than the population to which the numerator care is usually applied in regular practice. This is because we wanted a sample to whom the care is clearly applicable. However, this method not only fails to give proper credit to the provider who makes these important decisions of tailored medicine, but also lowers the observed quality score (i.e. adherence rate to the quality indicator), because subtracting the same number from both the numerator and denominator actually decreases the fraction when the fraction is less than 1 (for example, (8-1)/(10-1) = 7/9 = 0.78 is less than 8/10 = 0.8). Further, relatively small target populations can render quality scores statistically unstable, with large standard errors.
A second method of handling exceptions is to treat them as if the care had in fact been provided, provided that the reason for the exception is properly documented. In this way, atypical patients can be kept in the sample and the care they receive can be evaluated in the same way as for regular patients. Quality scores using this method tend to be more stable than those using the first method above, but patient eligibility for the quality indicator becomes heterogeneous, which may then dilute the link between the quality and the outcome by introducing noise into the biological process-outcome link. While acknowledging this limitation, we basically adopted this second method of handling exceptions in the present study. This can be seen in the fact that many of our quality indicators explicitly demand documentation of the reason why the indicated care was not provided (Table ).
The third way of handling exceptions is to reduce target adherence by the expected number of exception patients[30
]. If 5% of patients ostensibly applicable to a quality indicator are estimated to become exceptions, the quality target can be set at 95% instead of 100%. Given that identifying individual exceptions requires that the researcher abstracting the medical record has knowledge of treatment choices, which is occasionally difficult due to a lack of detailed documentation[31
], a simple reduction in target enables the judgment of individual cases to be avoided, and also thus the risk of errors in such judgment. Further, gaming with the system can be discouraged by labeling patients who did not receive the indicated care as exceptions, perhaps retrospectively. In comparison of quality scores across populations, however, national variations in the proportion of exception cases may introduce large random errors into the comparison.
The above clearly demonstrates the difficulty in handling exceptions. Even when considering only individual cases, we may well refrain from stating that the target quality score is 100%, or instead state outright that the target is slightly lower than 100%. Indeed, the UK pay-for-performance initiative (Quality and Outcomes Framework) gives full points even for quality scores less than 100%[32
], although they also allowed providers to limit the denominators reporting exceptions. At the least, quality information should be interpreted together with data regarding the nature and number of exceptions reported by providers[30