We make every effort to select a treatment according to scientific principles. We try to avoid anecdotal reports of therapeutic efficacy and risk because factors such as placebo effect and individual variation in reactions to treatment can cloud the interpretation of individual responses. To avert these confounding variables, we rely heavily on randomized, controlled trials of therapeutic approaches. To qualify as an appropriate study, patients must be assigned to treatment randomly; neither the patient nor the physician must know which treatment is being administered; outcomes must be important ones (deaths and disability rather than intermediate test results); outcomes must be measured and defined with precision; and analysis of data must be done using accepted methods.
Such trials are laborious, expensive, and subject to flaws, both in design and implementation, yet controlled studies have provided many invaluable therapeutic insights. Often, however, even the best of the randomized, controlled studies provide only an anchor point or a benchmark when it comes to selecting therapy for an individual patient. To the extent that a given patient differs notably from the individuals studied in a randomized trial, that patient’s response to the treatment also might well differ.
Patients can differ in many ways, including their age, sex, race, genetic makeup, severity of illness, and the stage at which their disease is encountered. In addition, physicians often encounter a clinical problem for which no randomized, controlled trial has been carried out. When the patient fails to match a cohort in a controlled trial or when no such trial is available, the physician’s judgment is the fallback position. The elements of therapeutic judgment become critical in such circumstances, forming the basis for the ability to evoke principles in making therapeutic decisions in the face of uncertainty.
Treatment Under Conditions of Uncertainty
( Case 41, Case 44, Case 48, Case 50, Case 53)
Therapeutic considerations frequently focus on the specific characteristics of one or another treatment, including the efficacy of a drug or of an interventional approach, and the risks of treatment. However, these important influences on choices of one therapeutic approach over another seldom occur in isolation. Treatment decisions often must be made before a diagnosis has been confirmed; furthermore, in some circumstances diagnostic uncertainty is never resolved, yet treatment decisions must be made. The principles described in Chapter 4 guide decision making under conditions of diagnostic uncertainty.
Restated briefly, they are as follows: When the efficacy of the available treatment for a given disease is low or if the risk of the treatment is unusually high (or both), the treatment should only be given if the probability of the disease is quite high. On the other hand, if the risk of the treatment is negligible or if the efficacy of therapy is unusually high, the treatment can be given even when the probability of the disease is quite low (see Fig. 4.10).[ 57],[ 58] Therapeutic implications of test use follow similar rules.
When the probability of disease is very high, a negative test result usually will not reduce the suspicion of disease sufficiently to change the original assessment of the need for treatment. If so, the test is unnecessary. When the probability of disease is very low, a positive test result often will not increase the suspicion of disease sufficiently to change the physician’s mind about the lack of a need for treatment. If so, the diagnostic test also is superfluous. However, if the test result can be expected to alter the probability of disease sufficiently to influence the decision to give or to withhold a treatment, then the test should be used.[ 57],[ 59]
When the Value of Therapeutic Choices is Close
( Case 25, Case 26, Case 46, Case 50, Case 51)
Unfortunately, the threshold determinations discussed in detail under diagnostic testing may not provide a definitive answer to the decision on whether to give or withhold a treatment. Just because a disease probability falls above or below a threshold, the differences in value between giving no treatment and giving treatment may be quite small and thus may be clinically insignificant.[ 60] When comparing the choice of withholding versus giving a treatment or when comparing two treatments, the clinician tries to assess the benefit of one approach over another. In many instances, this benefit is large and the decision is clear.
In some instances, however, no clear therapeutic approach dominates. A difference of only a few days in life expectancy between two choices may imply that the decision between the choices is so close that neither choice predominates. When two or more choices are imperceptibly different in their perceived values (or expected utility, in the language of decision theory), the decision is considered a close call, or a toss-up.[ 60] In such circumstances, minor differences in patients’ preferences may help to decide whether to give one treatment or another. When testing is one of the choices, a desire of the patient to know a test result may be sufficient to move the decision toward further testing.
The principal problem in dealing with therapeutic toss-ups lies in judging the clinical relevance of a small marginal benefit. A difference of several years of life expectancy between two treatments seems like quite a lot, whereas when the difference is only several days or weeks, the physician could easily recommend either treatment. However, even a difference of a few weeks could be important to a particular patient. Given these features of therapeutic decision making, patients’ preferences must always be taken into consideration. Doing so is especially important when differences in the outcomes of two choices are quite small.
( Case 46, Case 47)
Therapeutic decision making often involves making complex tradeoffs between choices that are not easily balanced against one another. In some instances, the choice may lie between one approach in which the risk of therapy is immediate and the expected beneficial effect of therapy is long term, and another approach that involves no immediate risk but with which there are important possible long-term unfavorable outcomes (e.g., the morbidity and mortality associated with immediate cholecystectomy for asymptomatic gallstones vs. the later enhanced risks of subsequent surgery for serious complications of gallstones at an older age).
In other instances, one must weigh the immediate effects of a particular therapy on morbidity and mortality versus the long-term effects of that therapy on the quality of a patient’s life (e.g., the risks in terms of morbidity and mortality of joint replacement for an arthritic hip vs. the long-term benefit of surgery in terms of improved mobility).
Quantitative Therapeutic Decision Making
( Case 23, Case 30, Case 45, Case 47, Case 51)
Many therapeutic decisions must be made before all diagnostic information is available and before we are confident of a diagnosis. In many instances, selection of therapy is simple and straightforward because extensive experience has confirmed the value and safety of a given approach. In such instances, we develop comfortable and familiar categorical rules of procedure (“treatments of choice”) that guide our decision making. Given the repetitiveness of our day-to-day patient experiences, this practice generally stands us in good stead. Nonetheless, situations often arise in which the patient or clinical setting is in some way atypical—the operative mortality may be higher than usual because of a patient’s risk factors and comorbid illnesses, there may be considerable diagnostic uncertainty, or the efficacies of alternative therapies may be in doubt.
Sometimes we are confronted with innovative techniques for testing or novel therapies, developments in health technology for which we do not yet have adequate information.[ 61] When these problems stretch the judgmental capacities of physicians, a quantitative approach to therapeutic decision making known as decision analysis can be used. Decision analysis applies probability and utility theory to therapeutic decision making under conditions of uncertainty.[ 10],[ 11]
The process requires structuring the therapeutic dilemma as a decision tree that contains all choices and outcomes, specifying the probability and the utility (value) of each outcome, and making a calculation from these data to determine the optimal choice. Given the quantitative nature of the data used in this decision-making process and the ease by which computerized decision trees can be recalculated, the data used in the analysis can be tested for its influence on the decision. The process by which the robustness of a decision is assessed by testing it against reasonable limits of the data is called sensitivity analysis. In sensitivity analysis, the effect of any single probability or sets of probabilities can be tested. The effect of utility values can be assessed in the same manner.
One can ask, for example, whether the decision would be different if the probability of a certain therapeutic response were higher or lower, or if the quality of life (i.e., the utility) of a given outcome were higher or lower. Because computer programs can carry out extensive calculations with combinations of probabilities and utilities, thresholds can be derived not only for the probabilities that affect a decision, but also for critical outcome measures (utilities). If necessary, the effect on the decision of variations in multiple variables can be appraised simultaneously.
Even though computer technology has greatly simplified construction and assessment of decision trees, decision analysis for complex clinical problems must be used with considerable caution by inexperienced individuals. Because such analyses are quite sensitive to decision tree structure and the data used in the analysis, it is better to leave such analyses to experts. Even experts use their common sense and clinical judgment in interpreting the results of decision analyses when the outcomes are counterintuitive. In such circumstances, the analysts scour their assumptions, check the structure of their decision models, and return to literature searches before assuming that their analysis is rational.