Early health technology assessment of machine learning-enabled remote monitoring for advanced Parkinson disease

Advanced Parkinson disease (APD) is often under-recognized, which may delay treatment optimization and referral for device-aided therapy assessment. Machine learning-enabled remote monitoring systems (ML-RMS) may support earlier recognition of progression, but their value depends on whether alerts lead to clinical action and patient benefit.

We developed a Markov model comparing standard care with ML-RMS-supported care from the Finnish health care perspective. The model simulated persons with Parkinson disease at risk of APD over 16 years. The base case assumed ML-RMS sensitivity and specificity of 88 percent and 62 percent. Costs, quality-adjusted life years (QALYs), incremental cost-effectiveness ratios, and incremental net monetary benefit were estimated. Scenario and probabilistic sensitivity analyses explored diagnostic accuracy, monitoring costs, direct quality-of-life effects, deep brain stimulation (DBS) costs, pathway timing, and delayed nursing home admission.

In the base case, ML-RMS generated 0.0044 additional QALYs at an additional cost of €761 per patient, resulting in an incremental cost-effectiveness ratio of €172,023 per QALY gained. At a willingness-to-pay threshold of €35,000 per QALY, the incremental net monetary benefit was negative. Scenario analyses indicated that ML-RMS could become cost-effective if active monitoring produced direct quality-of-life gains or delayed nursing home admission.

ML-RMS is unlikely to be economically favorable if limited to earlier APD recognition and timely DBS treatment in a small eligible subgroup. Its value depends on whether monitoring changes care pathways, improves patient outcomes, or reduces long-term care needs. Future studies should prioritize clinical utility, quality-of-life effects, treatment decisions, and institutional care.