Timing of repeated monitoring tests

This article describes a method for evaluating some of the medical, economic, and personal consequences of different policies of monitoring patients for various disorders. The problem is difficult because it is dynamic. Most conditions develop over time, and in many cases the sensitivity of the test as well as the cost of treatment and the prognosis of the patient depend on how far the disorder has progressed at the time the test is performed. This model incorporates all these factors and enables one to estimate the consequences of applying several tests in different orders and frequencies. The most important assumption in the model is that the outcomes are functions of the earliness of detection (as defined in Figure 3). Failure of this assumption can sometimes be corrected for, but the equations are considerably more complicated and the estimation of parameters more difficult. The progression assumption, that once a particular condition is detectable it is always detectable, is a strong assumption that does not admit the possibility of random false-negative test results. The presence of random false-negative results, however, can be tested for and a random false-negative rate can be included in the model if this is important. This model, like any model, is limited by how well the assumed functional forms (for example, the functional form of the occult interval) represent the actual functional relationships, and by the quality of the data used for estimation of parameters. In some cases, such as screening for breast cancer, good data are available and different functional forms can be tried. In the case of most monitoring tests, however, good data are not currently available. The strength of the model is that it can be used to analyze rather complicated problems. Its main role is to provide general insights, to identify what data should be collected, and to suggest how these data can be used.