Scholarlink

Background <p>Medical claims are widely used in workforce estimation and health services research but identifying primary care (PC) clinicians is challenging. Existing methods relying on specialty or activity-based metrics are imprecise and hard to duplicate.</p> Objective <p>We seek to develop a simple, accurate decision tree classifier using claims data, trained on survey responses from family physicians (FPs). And to use this classifier to estimate the PC workforce in Virginia.</p> Design <p>We linked 2016–2023 Virginia All-Payer Claims Database (APCD) data with responses from American Board of Family Medicine surveys to infer whether FPs provided PC. Using claims-derived features, we trained a decision tree to classify clinicians providing PC. We developed an enhanced version of the tree, adding exclusion criteria, and applied both classifiers to the entire APCD, including other PC clinician types.</p> Subjects <p>Virginia clinicians.</p> Main Measures <p>Estimated classifier accuracy by clinician type and workforce estimates with bootstrapping used to estimate 95% confidence intervals.</p> Key Results <p>The base decision tree correctly classified 93% of the training sample using three clinician-level features: percent of claims with diagnosis category of Z00, percent with place of service 19–23 or 31–32, and percent of patients having multiple visits with the clinician. We estimate the enhanced tree achieved at least 89% accuracy for every clinician type. We found that from 2016 to 2023, the percentage of Virginia PC clinicians who were nurse practitioners (NPs) grew from 17 to 32%, with NPs passing FPs as the most common PC clinician type.</p> Conclusions <p>This decision tree approach overcomes shortcomings of existing methods and offers a straightforward, scalable, interpretable tool for classifying PC clinicians, with applications in workforce planning and health services research.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Reclassifying Primary Care: A Decision Tree Approach to Improving Workforce Estimates and Research Using Claims Data

  • 期刊论文

  • 发表日期：

    2026年5月11日

查看全文
  • Zachary J. Morgan,
  • Andrew W. Bazemore,
  • Lars E. Peterson,
  • Mingliang Dai

摘要

Background

Medical claims are widely used in workforce estimation and health services research but identifying primary care (PC) clinicians is challenging. Existing methods relying on specialty or activity-based metrics are imprecise and hard to duplicate.

Objective

We seek to develop a simple, accurate decision tree classifier using claims data, trained on survey responses from family physicians (FPs). And to use this classifier to estimate the PC workforce in Virginia.

Design

We linked 2016–2023 Virginia All-Payer Claims Database (APCD) data with responses from American Board of Family Medicine surveys to infer whether FPs provided PC. Using claims-derived features, we trained a decision tree to classify clinicians providing PC. We developed an enhanced version of the tree, adding exclusion criteria, and applied both classifiers to the entire APCD, including other PC clinician types.

Subjects

Virginia clinicians.

Main Measures

Estimated classifier accuracy by clinician type and workforce estimates with bootstrapping used to estimate 95% confidence intervals.

Key Results

The base decision tree correctly classified 93% of the training sample using three clinician-level features: percent of claims with diagnosis category of Z00, percent with place of service 19–23 or 31–32, and percent of patients having multiple visits with the clinician. We estimate the enhanced tree achieved at least 89% accuracy for every clinician type. We found that from 2016 to 2023, the percentage of Virginia PC clinicians who were nurse practitioners (NPs) grew from 17 to 32%, with NPs passing FPs as the most common PC clinician type.

Conclusions

This decision tree approach overcomes shortcomings of existing methods and offers a straightforward, scalable, interpretable tool for classifying PC clinicians, with applications in workforce planning and health services research.