Scholarlink

Introduction <p>Developing countries like Tanzania rely heavily on out-of-pocket (OOP) payments to finance health. Unfortunately, these OOP payments often expose patients to catastrophic health expenditures (CHE), which hinders progress toward universal health coverage (UHC). Tanzania introduced an improved Community Health Fund (iCHF) in 2017/2018 as a prepayment insurance mechanism to reduce the burden of healthcare costs on its population. To our knowledge, there is limited literature in Tanzania using secondary data to predict iCHF membership retention rate. This study aimed to develop a machine learning model for predicting iCHF membership retention using 2019/2020 and 2020/2021 data.</p> Methods <p>This retrospective study targeted 4,020,750 beneficiaries registered in Tanzania’s iCHF health information management system. Claims data were retrieved from iCHF’s electronic database covering the period from 2019 to 2021. A logistic regression model was used as the modelling approach. A confusion matrix was used to calculate model performance metrics such as Area Under the Receiver Operating Characteristic Curve (AUC-ROC), accuracy, sensitivity, and specificity.</p> Results <p>Annual membership renewal rates in 2019/2020 and 2020/2021 were comparable at 71.5% and 74.8%, respectively. The logistic regression model produced an AUC-ROC of 0.61, indicating a slightly better performance than a random classifier with an AUC-ROC of 0.5. Using a 0.5 cut-off, the model had an accuracy of 71.4%, sensitivity of 99.8% and specificity of 2.6%. The model’s performance significantly improved by adjusting the prediction threshold from 0.5 to 0.7, which changed the accuracy to 60%, sensitivity to 62%, and specificity to 53%.</p> Conclusion <p>The findings demonstrate the potential of data mining and big data analytics techniques in health insurance data, which is highly underutilised in developing countries. The study has also demonstrated that machine learning models can be trained on iCHF data. Recommendations from this study provide vital considerations in the design of future databases so that they can generate high-quality secondary data.</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

Prediction of insurance membership retention rates using machine learning: a case study of Tanzania’s improved community health insurance fund (iCHF)

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月12日

查看全文
  • Castory Munishi,
  • George Ruhago,
  • Amani Mori,
  • Jofrey S. Amos,
  • Othman Haji,
  • Alphoncina Kagaigai,
  • James T. Kengia,
  • Oddvar Martin

摘要

Introduction

Developing countries like Tanzania rely heavily on out-of-pocket (OOP) payments to finance health. Unfortunately, these OOP payments often expose patients to catastrophic health expenditures (CHE), which hinders progress toward universal health coverage (UHC). Tanzania introduced an improved Community Health Fund (iCHF) in 2017/2018 as a prepayment insurance mechanism to reduce the burden of healthcare costs on its population. To our knowledge, there is limited literature in Tanzania using secondary data to predict iCHF membership retention rate. This study aimed to develop a machine learning model for predicting iCHF membership retention using 2019/2020 and 2020/2021 data.

Methods

This retrospective study targeted 4,020,750 beneficiaries registered in Tanzania’s iCHF health information management system. Claims data were retrieved from iCHF’s electronic database covering the period from 2019 to 2021. A logistic regression model was used as the modelling approach. A confusion matrix was used to calculate model performance metrics such as Area Under the Receiver Operating Characteristic Curve (AUC-ROC), accuracy, sensitivity, and specificity.

Results

Annual membership renewal rates in 2019/2020 and 2020/2021 were comparable at 71.5% and 74.8%, respectively. The logistic regression model produced an AUC-ROC of 0.61, indicating a slightly better performance than a random classifier with an AUC-ROC of 0.5. Using a 0.5 cut-off, the model had an accuracy of 71.4%, sensitivity of 99.8% and specificity of 2.6%. The model’s performance significantly improved by adjusting the prediction threshold from 0.5 to 0.7, which changed the accuracy to 60%, sensitivity to 62%, and specificity to 53%.

Conclusion

The findings demonstrate the potential of data mining and big data analytics techniques in health insurance data, which is highly underutilised in developing countries. The study has also demonstrated that machine learning models can be trained on iCHF data. Recommendations from this study provide vital considerations in the design of future databases so that they can generate high-quality secondary data.