Health Informatics Assignment Week 3

Questions for Discussion

1. Calculate the following probabilities for a patient about to undergo CABG surgery (see Example 2):
   1. The only possible, mutually exclusive outcomes of surgery are death, relief ofsymptoms (angina and dyspnea), and continuation of symptoms. The probability of death is 0.02, and the probability of relief of symptoms is 0.80. What is the probability that the patient will continue to have symptoms?
   2. Two known complications of heart surgery are stroke and heart attack, with prob-abilities of 0.02 and 0.05, respectively. The patient asks what chance he or she has of having both Assume that the complications are conditionally independent, and calculate your answer.
   3. The patient wants to know the probability that he or she will have a stroke giventhat he or she has a heart attack as a complication of the surgery. Assume that 1 in 500 patients has both complications, that the probability of heart attack is 0.05, and that the events are independent. Calculate your answer.
2. The results of a hypothetical study to measure test performance of the PCR test forHIV (see Example 1) are shown in the 2 × 2 table in Table 3.9.
   1. Calculate the sensitivity, specificity, disease prevalence, PV⁺, and PV^–.
   2. Use the TPR and TNR calculated in part (a) to fill in the 2 × 2 table in Table 3.10. Calculate the disease prevalence, PV⁺, and PV^–.
3. You are asked to interpret a PCR HIV test in an asymptomatic man whose test waspositive when he volunteered to donate blood. After taking his history, you learn that he is an intravenous-drug user. You know that the overall prevalence of HIV infection in your community is 1 in 500 and that the prevalence in intravenous-drug users is 20 times as high as in the community at large.
   1. Estimate the pretest probability that this man is infected with HIV.
   2. The man tells you that two people with whom he shared needles subsequently diedof AIDS. Which heuristic will be useful in making a subjective adjustment to the pretest probability in part (a)?

      Table 3.9. A 2 × 2 contingency table for the hypothetical study in problem 2.

      PCR test result	Gold standard test positive	Gold standard test negative	Total
      Positive PCR	48	8	56
      Negative PCR	2	47	49
      Total	50	55	105
      PCR = polymerase chain	reaction.
      Table 3.10. A 2 × 2 co	ntingency table to complete for p	roblem 2b.
      PCR test result	Gold standard test positive	Gold standard test negative	Total
      Positive PCR	x	x	x
      Negative PCR	100	99,900	x
      Total	x	x	x

      PCR = polymerase chain reaction.

   3. Use the sensitivity and specificity that you worked out in 2(a) to calculate the post-test probability of the patient having HIV after a positive and negative test. Assume that the pretest probability is 0.10.
   4. If you wanted to increase the post-test probability of disease given a positive testresult, would you change the TPR or TNR of the test?
4. You have a patient with cancer who has a choice between surgery or chemotherapy.If the patient chooses surgery, he or she has a 2 percent chance of dying from the operation (life expectancy = 0), a 50 percent chance of being cured (life expectancy = 15 years), and a 48 percent chance of not being cured (life expectancy = 1 year). If the patient chooses chemotherapy, he or she has a 5 percent chance of death (life expectancy = 0), a 65 percent chance of cure (life expectancy = 15 years), and a 30 percent chance that the cancer will be slowed but not cured (life expectancy = 2 years). Create a decision tree. Calculate the expected value of each option in terms of life expectancy.
5. You are concerned that a patient with a sore throat has a bacterial infection thatwould require antibiotic therapy (as opposed to a viral infection, for which no treatment is available). Your treatment threshold is 0.4, and based on the examination you estimate the probability of bacterial infection as 0.8. A test is available (TPR = 0.75, TNR = 0.85) that indicates the presence or absence of bacterial infection. Should you perform the test? Explain your reasoning. How would your analysis change if the test were extremely costly or involved a significant risk to the patient?
6. What are the three kinds of bias that can influence measurement of test performance?Explain what each one is, and state how you would adjust the post-test probability to compensate for each.
7. How could a computer system ease the task of performing a complex decision analysis? Look at the titles of Chapters 9 through 18 of this text. What role could each kind of system play in the medical-decision process?
8. When you search the medical literature to find probabilities for patients similar to oneyou are treating, what is the most important question to consider? How should you adjust probabilities in light of the answer to this question?
9. Why do you think physicians sometimes order tests even if the results will not affecttheir management of the patient? Do you think the reasons that you identify are valid? Are they valid in only certain situations? Explain your answers. See the January 1998 issue of Medical Decision Making for articles that discuss this question.
10. Explain the differences in three approaches to assessing patients’ preferences forhealth states: the standard gamble, the time trade-off, and the visual analog scale.

Answer

1. 1. If x is the probability that the symptoms will continue, then:

      [(probability of death) + (probability of relief from symptoms)] = x

      [(0.02) + (0.80)] = x

      Therefore, x= 0.18

   2. The probability of both the events occurring independently:

      p [A, B] = p (A) * p (B)

      So, p [heart stroke, heart attack] = p (heart stroke) * p (heart attack)

      p [heart stroke, heart attack] = (0.02) * (0.05)

      p [heart stroke, heart attack] = 0.001

   3. p [stroke] [heart attack] = p [both the conditions]

      x * 0.05 = 0.002

      x = 0.002/ 0.05

      x = 0.04

2. 1. Sensitivity= 0.96 (48/50)

      Specificity= 0.85 (47/55)

      Disease prevalence = 50/105 x 100% = 47%

      TPR= 85% (48/56 x 100)

      TNR= 89% (49/55 x 100)

3. Disease prevalence=

PV+ =

PV- =

6. The word 'distortion' refers to a systemic bias of the interaction among care, risk factor or experience on the one side and clinical outcomes on another. One may discern three types of bias: knowledge bias, response bias and confounding. In a randomized analysis of muscle pain headache clients, the general practitioner associated pain management to routine treatment. One of the results of this study, as reported by the participants, was 'headache severity.' There has been no momentarily blinded: the patients have understood that they receive pain management or routine treatment. This could have given rise to 'wanted' reactions, particularly in the group that received pain management. This would result in inaccurate (too low) calculation of headache incidence throughout this category, resulting in as well favorable an estimation of its impact of therapeutic exercise on headache intensity. This is an illustration of a bias to knowledge. Information bias is a systemic distortion of the correlation between determining factor and consequence by using inaccurate information about determining factor or consequence (or even both). The most prominent example is inaccurate determinant or result calculation. If the result standard errors focus upon this determining factor, or the determinant's standard errors relies mostly on result, that is usually marked as differential misclassification. Selection bias is the bias created by choosing people, groups or information for study so as to not ensure satisfactory random sampling, therefore guaranteeing that the study sample is really not representative of the general population to be investigated. The distribution impact is often pointed to as this. The selection bias can be avoided by employing these measures like usage of random strategies when choosing individuals from subpopulation and ensuring the chosen subsets are similar to the general people in general of certain main features. Confounding tends to occur whenever an element is connected both to the determining factor as well as the actual result and destroys the possible correlation between both the 2. In principle the confounding factor (the confounder) is both the determinant 's cause as well as the result's source. Factors resulting from a determining factor cause and result are often seen as confusing in implementation. For analysis into a causal association among determinant and effect, confounding is indeed a limitation that may emerge. Confounding can indeed be eliminated in experiments into the results of a medication by administering the drug to the respondents in a sample by likelihood: a clinical study. For this case, the study groups are supposed to still be equivalent for the result in calculated and unknown chronic diseases.