Ensuring the quality of data is a critical step in data analysis. Our data scientists are proficient in reviewing primary data sources and determining whether they can be validated, supplemented, or replaced by existing administrative records. Additionally, our data governance experts examine and improve organizational processes that enhance the creation, transformation, and use of data for analytical purposes.
Missing data and data that are input incorrectly are common in any large-scale survey data set. Incorrect data need to be identified and removed before analysis can take place. The Data Science and Advanced Analytics team worked with researchers at AIR to assess the pattern of responses to the CRDC. In particular, the team identified problematic data points using an outlier and anomaly detection model. This model identified potential errors with far greater speed and efficiency than manual review. Results were used to inform decisions prior to data collection as well as business rules and data editing practices.
The Medicare Part D Enhanced Medication Therapy Management (Enhanced MTM) model tests whether providing Part D sponsors with additional payment incentives and regulatory flexibilities will lead to improved therapeutic outcomes and reduce net Medicare expenditures. Supporting documents submitted by participating sponsors allow the Centers for Medicare & Medicaid Services (CMS) to validate that Enhanced MTM encounter records reflect real-world events. These documents are manually reviewed each quarter for a sample of encounter records, including call logs, notification documents, records of recommendations, evidence that prescribers accepted or rejected recommendations, and all other documents appropriate to support the encounter codes reported in the records.
The DSAA team developed an integrated set of tools to automate the manual review process, from data extraction to data field validation, for about 11,000 supporting documents accumulated from six Enhanced MTM sponsors. This automated solution considerably accelerated the review process for the documents.
The Bundled Payments for Care Improvement Advanced (BPCI Advanced) Model is part of efforts by the Centers for Medicare & Medicaid Services (CMS) to implement voluntary episode payment models. To better understand model participants (e.g., hospitals) and their infrastructure readiness, staff training, program compliance, and reporting procedures, an Annual Check-in Questionnaire (ACQ) was developed to collect data from BPCI Advanced participants. The ACQ consists of about 80 questions in numerous formats such as drop-down menu-based multiple-choice questions, checklists, radio buttons, and an open-ended text input box. The ACQs were received from more than 500 participants.
Manually processing the ACQs can be time consuming and prone to errors. To accelerate this process, the Data Science and Advanced Analytics (DSAA) team used Python to develop an automation tool that quickly extracts questionnaire data and validates ACQ submissions by identifying corrupted files. The tool also automatically computes internal risk scores based on participants’ responses to questions that are indicative of risk of noncompliance with program rules. The tool also features a user-friendly graphic interface that allows users to perform data extraction and validation in a few clicks.
