model in particular and about the more general theoretical foundation for attempting to measure mortality as a hospitalwide characteristic, given how hospital-specific variations in end-of-life care can influence such measures (Fisher et al., 1994). The NQCB will need to address these issues as the measurement development effort goes forward.
Shared accountability is cross-cutting in that it holds all providers who partake in a patient’s care responsible for the outcomes of that care. There is no single method for achieving shared accountability. Assessing care longitudinally across time and space can require the evaluation of care for a patient from the hospital to the nursing home. Composite measures of care reinforce this overall approach by focusing on treatment for all aspects of a patient’s condition. Measurement at both the population and systems levels addresses the larger health care system and includes the societal factors that contribute to the health of the general public. The committee therefore believes that development and promulgation of measures in all of these areas foster shared accountability. It will become increasingly necessary to develop models of shared accountability as the focus shifts away from measuring care by setting, as discussed in the next section.
The NQCB should support research aimed at resolving key methodological issues surrounding performance measurement so as to enhance the accuracy and integrity of the data obtained. If measurement methodologies are flawed, data can be misleading, potentially threatening providers’ reputations and falsely portraying the quality of care provided. The committee calls particular attention to the following issues:
Risk adjustment—This statistical tool allows data to be modified to control for variations in patient populations. For example, risk adjustment could be used to ensure a fair comparison of the performance of two providers: one whose caseload consists mainly of elderly patients with multiple chronic conditions and another who treats a patient population with a less severe case mix. Risk adjustment makes it possible to take these differences into account when resource use and health outcomes are compared.
Sample size—Small sample sizes may make conclusions statistically invalid, particularly when used for ranking individual providers. For instance, depictions of a physician’s performance may be inaccurate if she has