The Problem
Peripheral artery disease (PAD) remains underdiagnosed and undertreated. PAD patients suffer an increased risk of heart attack, stroke, amputation, and death, with minority and lower socioeconomic status patients more affected. Untreated Duke patients with PAD have a 49% hospitalization rate, 8% amputation rate, and 15% mortality rate, which is at the poorer end of outcomes nationally.1,2 These poor clinical outcomes are associated with increased costs to patients and our health system. At Duke, there is an opportunity to expand care for this disease through population health monitoring of PAD.
Our Solution
Our team is improving the care of Duke patients with PAD through proactive identification and intervention to address gaps in their care.
We designed and implemented a PAD virtual rounding program that includes a PAD predictive model and generates intervention recommendations to patients’ primary care physicians (PCPs) ahead of scheduled visits. Our goal is to reduce downstream costs and improve equity and outcomes of PAD care within our health system.
We implemented a previously developed and validated logistic regression machine learning model that successfully identified PAD patients based on diagnosis codes and another encounter history within the Electronic Health Record (EHR).3 Our model runs on all adult patients with a Duke University Health System (DUHS) clinical encounter from the start of the EHR that includes at least one of 108 PAD-related diagnosis codes. A risk score is generated for each patient in the model cohort, indicating the likelihood that the patient has PAD. For our pilot, we focus on patients with primary care provider (PCP) appointments in the upcoming week to provide an actionable recommendation to the PCP at a time when they are thinking about the patient.
Patients above the threshold are put into a rounding cohort, then assessed by a PAD specialist for inclusion in the virtual rounds discussion. The specialist verifies the patient’s PAD disease status and whether the patient would benefit from an intervention, most commonly lipid-lowering medication adjustment (statins and PCSK9 inhibitors), and smoking cessation clinic referrals. Final recommendations are discussed weekly on a multidisciplinary team rounds discussion that includes the PAD specialist, a pharmacist, and a Duke population health lead. Recommended changes to care are communicated to the patient’s PCP, who can then tailor the patient’s care plan at their upcoming appointment. The model runs on a weekly interval to identify new patients and update scores for existing patients. Figure 1 illustrates the workflow of these ongoing rounds.
Impact
Starting in January 2022, we successfully implemented an ongoing population health rounding intervention for PAD patients that identifies PAD patients within the DUHS. Patients with a model-generated risk score above the threshold enter the rounding cohort, where they are discussed during weekly PAD population-level rounds. In the initial six months of rounds, 237 patients entered active rounds discussions, 56 were recommended for medication adjustments, and 45 for smoking cessation referrals. Other interventions included ankle-brachial index assessment and care management referral.
Next Steps
We will evaluate the impact of our pilot on clinical and cost outcomes for our patients, as compared with Duke patient population pre-implementation. We will continue to monitor model accuracy and will assess an alternative natural language processing model, which was previously shown to identify PAD patients with higher accuracy. We will maintain the PAD rounding process, seeking feedback for optimization from key stakeholders, especially from our PCPs. We aim to expand similar virtual rounds-based workflows for PAD at other institutions.
Academic Output
Improving Equity and Value of Peripheral Artery Disease Care at a Population Level. Machine Learning for Healthcare 2022 – Clinical Abstract Software and Demo Track. August 5-6, 2022. Rebecca Shen, E. Hope Weissler, William Ratliff, Marshall Nichols, Bradley Hintze, Michael Gao, Pamela Cohen, Holly Alvarado, Dennis Narcisse, Mary Schilder, Tara Kinard, Benjamin Smith, Daniel Costello, Steven Lippmann, Mark Sendak, Suresh Balu, Schuyler Jones.
References
- Kalbaugh CA, Loehr L, Wruck L, et al. Frequency of care and mortality following an incident diagnosis of peripheral artery disease in the inpatient or outpatient setting: the ARIC (Atherosclerosis Risk in Communities) study. J Am Heart Assoc 2018;7(8):e007332. https://doi.org/10.1097/PCC.0000000000000254
- Weissler EH, Ford CB, Narcisse DI, Lippmann SJ, Smerek MM, Greiner MA, Hardy NC, O’Brien B, Sullivan RC, Brock AJ, Long C, Curtis LH, Patel MR, Jones WS. Clinician Specialty, Access to Care, and Outcomes Among Patients with Peripheral Artery Disease. Am J Med. 2022 Feb;135(2):219-227. doi: 10.1016/j.amjmed.2021.08.025. Epub 2021 Oct 7. PMID: 34627781; PMCID: PMC8840959. https://pubmed.ncbi.nlm.nih.gov/34627781/
- Weissler EH, Lippmann SJ, Smerek MM, Ward RA, Kansal A, Brock A, Sullivan RC, Long C, Patel MR, Greiner MA, Hardy NC, Curtis LH, Jones WS. Model-Based Algorithms for Detecting Peripheral Artery Disease Using Administrative Data From an Electronic Health Record Data System: Algorithm Development Study. JMIR Med Inform. 2020 Aug 19;8(8):e18542. DOI: 10.2196/18542. PMID: 32663152; PMCID: PMC7468640. https://pubmed.ncbi.nlm.nih.gov/32663152/
