Risk Assessment 

This lecture we focus on patients’ prognosis / risk / severity of illness. Not all patients are the same. They may have the same disease but radically different severity of illness. All analysis of healthcare outcomes starts with understanding the patient and their comorbidities better. This week we answer age old patient questions such as “What is going to happen to me?” A few years ago, I designed the multimorbidity indices. These indices are quite accurate and a good way to capture differences among the patients. We will discuss how CMS currently measures patients’ prognosis. The information you learn this week will be used in subsequent weeks to do riskadjusted control charts. This week we focus on the use of these indices in measuring patients’ prognoses. . Assigned Reading
Presentations
AssignmentsInstruction for Submission of Assignments: Assignments should be submitted directly on Blackboard. In rare situations assignments can be sent directly by email to the instructor. Submission should follow these rules:
Question 1: Construct a simple MultiMorbidity Index. Assess the average severity of CHF, MI, Diabetes, Hypertension, Alcohol Use, and ACL surgery (assume that sicker patients have longer stays). To calculate the average severity associated with a disease, compare all cases with the disease to all control patients without the disease. Make sure that in each comparison, patients with and without disease have the same set of comorbidities. For example, to find the average length of stay for patients with MI, select all MI patients with the following comorbidities: CHF,DM,AA. Then compare these cases to controls who do not have MI but have the same comorbidities. To help you understand this assignment, consider the following table. In this table, we see different strata of mutually exclusive and exhaustive comorbidities. Then, within each strata we can observe the impact of MI. The impact of MI is the average impact of MI within each strata. Your objective is to create this table before you calculate the impact of MI. To do so, first estimate n1 through n5 through a file where the data is restricted to cases with MI (WHERE MI=1). Then, estimate the values n6 through n10 through a file where the data are restricted to nonMI patients (WHERE MI=0). Merge these data, making sure that you match on the strata. Then you can calculate the impact of MI.
Question 2: Calculate the likelihood ratio associated with each diagnosis in predicting above or below average length of stay. First calculate the weighted average length of stay for each diagnosis. Assign individuals who have above weighted average length of stay 1 and those below 0. Next calculate the likelihood ratio (remembering to weight each strata by the count of patients that fall into each strata). To calculate the likelihood ratio, select all individuals who have above weighted average length of stay. Examine the prevalence of the diagnosis among them. Select all individuals who have below weighted average length of stay and select the prevalence of the diagnosis among them. The ratio of these two calculated numbers constitutes the likelihood ratio. Enter the likelihood ratios calculated for each diagnosis into a Table called #LR.
Question 3: Using the Table #LR (see question 2), calculate the probability of long stay (not length of stay but the probability of long stays) for the 3 patients described below. Note that diagnoses codes that do not have an associated likelihood ratio should be assigned a likelihood ratio of 1 or ignored. Please use SQL/Python code to calculate multiplication of the likelihood ratios. Since the data are presented in one column, the attached slides and video can help you calculate the probability of a set of likelihood ratios in one column. You should not enter the values by hand. You will not be able to do so when there are large number of patients.
Question 4: Calculate the likelihood ratio associated with each diagnosis. You would need to use SQL to do this assignment. This is a massive database of 17 million records. Microsoft SQL server can analyze the entire data in one run. Submit your SQL code and the 10 diagnosis with highest and lowest likelihood ratios. By opening this file you agree not to share the file with anyone else.
Question 5: Calculate the Receiver Operating Characteristic curve associated with predicting from age whether the patient will live or die. The data do not provide a predicted values, but you can construct a prediction based on age exceeding the cutoff value. If the patient's age exceeds the cutoff value, then we would predict that the patient will die. For example, the cutoff values for the ages provided can be 40, 50, 60, 70, 80, 90. If we take the cutoff value 60, then we predict that all patients above 60 will die and all patients below 60 will live. Calculate sensitivity and specificity for all cutoffs. Then, draw the Receiver Operating Characteristic curve. At what age cutoff the sum of the sensitivity and specificity is at its maximum?

Copyright © 1996 Farrokh Alemi, Ph.D. Created on Sunday, October 06, 1996 4:20:30 PM Most recent revision 09/21/2022. This page is part of the course on Quality, the lecture on Risk Assessment.
