This is a text written by Jon Berg <jon.berg|a|turtlemeat.com> spring 2005 in the Computer Science course Medical Informatics at Tromsø University, Norway.
Current and future possibilities of Medical Informatics
4. Medical Decision Making
Practicing medicine involves making decisions that involves uncertainty. For a doctor to make a decision that will affect the patient he must consider several sides of the case. There are often risks involved with any decisions. For example a person with hart problems may risk a hart attack if he is not operated, but also the surgery involves a risk, third there is also the risk of the surgery not having the desired effect and will cure the patient. The decision that the doctor makes is often based on a combination of deductive reasoning and previous knowledge collected on similar cases. The doctor can never be 100% sure that the decision he makes is the correct one.
To give the diagnosis the doctor performs a series of iterations of hypothesis generation, data collection and interpretations. This is called the hypothetico-deductive approach. For each of the iteration a more specific test gets performed to try to under build the case for that specific diagnosis. For each of the iteration the uncertainty of the wrong diagnosis gets smaller. The doctor also has to make a decision on what examination to do next in each of these iterations. There may be different cost to performing different the tests. The tests may also differ in the amount of more certainty they increase the diagnosis by. If the cost is too high or the certainty increase is too little it may not be worth to do the next test.
Probability is the preferred way to express uncertainty. It is expressed as a number between 0 and 1. Most assessments that physicians make are based on personal experience. To make an assessment based on a number of unconscious mental processes is called cognitive heuristics. For objective probability estimates a physician can use research results for reference. By placing a patient in a subgroup by the symptoms he has there is a possibility to look up the probability of a disease for a person in that group.
In diagnostics tests there are points where values of what can be considered healthy or diseased overlap. This introduces the possibilities of the groups:
A true positive: correctly classifies the patient as having the disease.
A true negative: correctly classifies the patient as not having the disease.
A false positive: the test incorrectly classifies the patient of having the disease.
A false negative: the test incorrectly classifies the patient as not having the disease.’
The rate of the likelihood of getting these groups can be controlled by choosing different cutoff values. It is a balance that must be taken according to what disease is diagnosed. It is always a trade off. Decreasing the number of false positives, would increase the number of false negatives. The type of disease and treatments must come into the consideration what values to use. If a disease is relatively harmless and the treatment is dangerous it is desirable to have few false positives. If a disease is life threatening and there is an available method to cure the disease it is desirable to have few false negatives.
The third stage of the diagnostic process is to adjust the probability estimate by using the information gained from the diagnostic tests and pre-test. Bayes’ theorem is a quantitative method for calculating the post-test probability using the pretest probability and the sensitivity and the specificity of the test. There can be problems with using the Bayes’ theorem in cases where there are; inaccurate estimation of pretest probability, faulty application of test performance measures or violations of the assumptions of conditional independence and mutual exclusivity. Another estimation method is the Positive predictive value; this method gives likelihood for a patient with a positive test also has the disease.
Decision Making Models
Often there are many possible decisions that can be made in choosing a treatment for a patient. The method of Expected-value decision making pins down each choice as a number and use the numbers as a way to find the best decision. The decision tree is used to visualize the different outcomes with probabilities associated with each possible outcome. Also quality-adjusted life (QALY) years are taken into consideration. QALY is a measurement for converting the value of years with poor health into years in fully functioning years. Sensitivity analysis can then be used to test if the conclusion in the analysis changes when the outcome estimates change within reasonable ranges. The decision tree is useful for decisions that will happen in the near future. A Markov model is better to model events that will happen over the lifetime for the patient. A doctor often has to make the decision of whether to treat a patient and not, this can be modeled with the treatment-threshold probability. Another way to model the decision process is influence diagrams.
Probability and decision analysis are tools to help the practitioner mange risk and help explain uncertainty to a patient. In some situations in may be sufficient to draw a possibility tree. In other situations it may be useful to do a rough estimate of the chances for the likely outcomes. In all situations most decisions must be made on imperfect data. Medical decision making is complicated there exists tools to give support in the process for example to draw decision trees, calculate expected values and perform sensitivity analysis. MEDLINE is a retrieval system that can be used in medical decision process for looking up estimates of disease prevalence and test performance.
Web (webmastering) Mix Computer related text
Mix Computer related text