[论文解读] Forecasts of Cancer and Chronic Patients: Big Data Metrics of Population Health

Jacob Kuriyan, Nathaniel Cobb|arXiv (Cornell University)|Jul 12, 2013

Nutritional Studies and Diet参考文献 11被引用 5

一句话总结

本文提出一种基于大数据的动态系统模型，可在12个月内以3%–6%的误差率预测癌症和慢性病发病率，实现对预防性医疗项目的效果实时评估。该研究引入一种衡量短期健康改善与成本节约的新指标，揭示所有研究的慢性病均显著增加癌症风险，其影响程度与糖尿病相似。

ABSTRACT

Chronic diseases and cancer account for over 75 percent of healthcare costs in the US. Increased prevention services and improved primary care are thought to decrease costs. Current models for detecting changes in the health of populations are cumbersome and expensive, and are not sensitive in the short term. In this paper we model population health as a dynamical system to predict the time evolution of the new diagnosis of chronic diseases and cancer. This provides a reliable forecasting tool and a means of measuring short-term changes in the health status of the population resulting from preventive care programs. Twelve month forecasts of cancer and chronic populations were accurate with errors lying between 3 percent and 6 percent. We confirmed what other studies have demonstrated that diabetes patients are at increased cancer risk but, interestingly, we also discovered that all of the studied chronic conditions increased cancer risk just as diabetes did, and by a similar amount. The model(i)yields a new metric for measuring performance of preventive and clinical care programs that can provide timely feedback for quality improvement programs;(ii)helps understand "savings" in the context of preventive care programs and explains how they can be calculated in the short term, even though they materialize only in the long term and(iii)provides an analytic tool and metrics to infer correlations and derive insights on the effect of changes in socio-economic factors affecting population health on improving health and lowering costs of populations.

研究动机与目标

• 利用人群健康指标开发一种数据驱动的慢性病与癌症发病率预测模型。
• 解决当前人群健康监测工具存在的缺陷，如反应迟缓、成本高昂且对短期变化不敏感。
• 创建一种新的预防性与临床医疗项目绩效评估指标，提供及时反馈以支持质量改进。
• 量化预防性医疗带来的短期‘节约’，尽管长期成本降低需经时间显现。
• 通过预测建模分析社会经济因素如何影响人群健康与医疗成本。

提出的方法

• 利用纵向大数据（包括已诊断的慢性病与癌症）将人群健康建模为动态系统。
• 应用时间序列预测技术，基于历史发病率趋势，预测未来12个月的新发病例。
• 采用统计验证方法评估预测准确性，报告癌症与慢性病人群的预测误差在3%至6%之间。
• 通过多变量分析评估既存慢性病与后续癌症风险之间的关联。
• 基于预测发病率趋势的偏离情况，推导出一种新的预防性医疗项目绩效评估指标。
• 将社会经济变量整合至模型中，以推断其对人群健康轨迹与成本结果的影响。

实验结果

研究问题

• RQ1 能否利用大数据，通过动态系统模型在12个月内准确预测新发癌症与慢性病病例？
• RQ2 如糖尿病等各类慢性病在多大程度上增加后续癌症诊断的风险？
• RQ3 如何利用基于预测的绩效指标实现实时评估预防性医疗项目？
• RQ4 即使成本节约需长期才能体现，预防性干预带来的短期健康改善应如何量化？
• RQ5 社会经济因素如何影响人群健康的演变轨迹及相应的医疗成本？

主要发现

• 12个月的癌症与慢性病发病率预测误差在3%至6%之间，表明预测具有高度准确性。
• 所有研究的慢性病（包括糖尿病）均以相似程度增加后续癌症诊断风险，提示存在广泛的生物学或行为学关联。
• 该模型提供了一种新的、可操作的绩效评估指标，可实现及时反馈，支持快速的质量改进循环。
• 通过检测发病率趋势与预测值之间的早期偏差，该模型可实现对预防性医疗‘节约’的短期量化。
• 该框架可通过预测分析推断社会经济变化对长期人群健康与医疗成本的影响。
• 本研究证实了既往关于糖尿病与癌症风险关联的研究发现，但进一步拓展了结论，表明风险升高并非仅见于糖尿病，而是广泛存在于多种慢性病中。