Data Mining and Predictive Analytics Flashcards

A Medicare Advantage plan analyzes its claims and pharmacy data to identify members with a diagnosis of diabetes who are also prescribed medications for chronic kidney disease (CKD), but have no corresponding diagnosis of CKD in their claims history. What is this data analysis process an example of?

Answer: C Predictive modeling for diagnostic suspecting.

This scenario describes predictive modeling, or suspecting, where data is mined to find patterns (e.g., specific prescriptions) that suggest the presence of an undocumented or uncoded chronic condition. The goal is to identify potential gaps in clinical documentation to be reviewed for accuracy, not to assume fraud or conduct a formal audit. [10, 6]

What is the primary purpose of using predictive analytics in a risk adjustment program?

Answer: C To identify and prioritize members with the highest likelihood of having undocumented, risk-adjusting conditions for review.

The core function of predictive analytics in risk adjustment is to efficiently sift through vast amounts of data to forecast which members are most likely to have active, but uncaptured, HCCs. This allows the organization to prioritize its resources, such as chart reviews and provider queries, on the cases with the highest probability of yielding accurate and complete coding. [4, 6, 12]

A health plan uses a data mining technique that groups members into distinct segments based on shared characteristics like prescription drug usage, frequency of specialist visits, and documented lab values. This helps them tailor different care management outreach programs. Which of the following techniques is being described?

Answer: D Cluster analysis

Cluster analysis is an unsupervised data mining technique used to partition a dataset into groups (clusters) where members in the same cluster are more similar to each other than to those in other clusters. This is commonly used in population health to identify patient cohorts with similar clinical profiles for targeted interventions. [1, 3, 14]

A risk adjustment analyst runs a model using pharmacy data that flags patients prescribed both insulin and gabapentin (often used for diabetic neuropathy) but who have no corresponding diagnosis code for diabetic neuropathy. The output of this model is primarily used to:

Answer: B Generate a list of members whose charts should be targeted for review to validate if the condition is documented.

The output of a predictive model is a lead, not a confirmed diagnosis. The compliant and logical next step is to use this information to investigate further. This involves a targeted review of the patient's medical record to see if the suspected condition is clinically documented. If it is, but was not coded, then it can be submitted. If it isn't, no action is taken on the code itself. [10]

In the context of data mining for risk adjustment, what is a significant limitation of relying SOLELY on claims data?

Answer: A It lacks the clinical detail and context found in unstructured medical record notes.

Claims data contains structured information like diagnosis and procedure codes but often lacks the rich clinical narrative, test results, physician reasoning, and symptomatology found in unstructured progress notes. This clinical nuance is often crucial for validating or identifying complex diagnoses, making claims data alone an incomplete picture for analysis. [2, 5, 7]

Which of the following data sources is LEAST likely to be a primary input for a predictive model designed to identify members with potentially undocumented risk-adjusting conditions?

Answer: B Patient satisfaction survey results.

Predictive models for risk adjustment rely on clinical and utilization data to predict the presence of disease. Pharmacy data, lab results, and historical claims are all direct indicators of a patient's health status. Patient satisfaction surveys measure a patient's experience and perception of care, which is not a direct clinical indicator of specific chronic diseases. [10]