A cluster-based approach for integrating clinical management of Medicare beneficiaries with multiple chronic conditions

BackgroundApproximately 28% of adults have ≥3 chronic conditions (CCs), accounting for two-thirds of U.S. healthcare costs, and often having suboptimal outcomes. Despite Institute of Medicine recommendations in 2001 to integrate guidelines for multiple CCs, progress is minimal. The vast number of unique combinations of CCs may limit progress.Methods and findingsTo determine whether major CCs segregate differentially in limited groups, electronic health record and Medicare paid claims data were examined in one accountable care organization with 44,645 Medicare beneficiaries continuously enrolled throughout 2015. CCs predicting clinical outcomes were obtained from diagnostic codes. Agglomerative hierarchical clustering defined 13 groups having similar within group patterns of CCs and named for the most common CC. Two groups, congestive heart failure (CHF) and kidney disease (CKD), included 23% of beneficiaries with a very high CC burden (10.5 and 8.1 CCs/beneficiary, respectively). Five groups with 54% of beneficiaries had a high CC burden ranging from 7.1 to 5.9 (descending order: neurological, diabetes, cancer, cardiovascular, chronic pulmonary). Six groups with 23% of beneficiaries had an intermediate-low CC burden ranging from 4.7 to 0.4 (behavioral health, obesity, osteoarthritis, hypertension, hyperlipidemia, ‘other’). Hypertension and hyperlipidemia were common across groups, whereas 80% of CHF segregated to the CHF group, 85% of CKD to CKD and CHF groups, 82% of cancer to Cancer, CHF, and CKD groups, and 85% of neurological disorders to Neuro, CHF, and CKD groups. Behavioral health diagnoses were common only in groups with a high CC burden. The number of CCs/beneficiary explained 36% of the variance (R2 = 0.36) in claims paid/beneficiary.ConclusionsIdentifying a limited number of groups with high burdens of CCs that disproportionately drive costs may help inform a practical number of integrated guidelines and resources required for comprehensive management. Cluster informed guideline integration may improve care quality and outcomes, while reducing costs.

背景：大约28%的成年人患有≥3种慢性疾病（CCs），这占美国医疗保健成本的三分之二，且往往导致不理想的结局。尽管美国医学研究所于2001年提出建议，要求整合针对多种CCs的指南，但进展甚微。大量独特的CCs组合可能限制了进展。方法与发现：为了确定主要CCs是否在有限的群体中存在差异性的分离，研究人员对一家有44,645名连续参保的Medicare受益人的电子健康记录和Medicare付费索赔数据进行了考察。通过诊断代码获取预测临床结局的CCs。使用聚合分层聚类法将具有相似组内CCs模式（以最常见的CC命名）的13个群体进行了划分。其中两个群体，充血性心力衰竭（CHF）和肾脏疾病（CKD），包括了23%的受益人，他们具有非常高的CC负担（分别达到10.5和8.1 CCs/受益人）。五个群体中，54%的受益人具有从7.1到5.9（按降序排列：神经、糖尿病、癌症、心血管、慢性肺病）的高CC负担。六个群体中，23%的受益人具有从4.7到0.4（按降序排列：行为健康、肥胖、骨关节炎、高血压、高脂血症、其他）的中低CC负担。高血压和高脂血症在各个群体中都很常见，而80%的CHF分离到CHF群体，85%的CKD分离到CKD和CHF群体，82%的癌症分离到癌症、CHF和CKD群体，85%的神经疾病分离到神经、CHF和CKD群体。仅在具有高CC负担的群体中，行为健康诊断较为常见。受益人/CCs的数量解释了索赔/受益人之间36%的方差（R²=0.36）。结论：识别出少量具有高CC负担的群体，这些群体不成比例地推动了成本，可能有助于确定所需的综合管理整合指南和资源数量。基于聚类信息的指南整合可能提高护理质量和结局，同时降低成本。