Smarter Medicine: The Future of Personal Health Care

The scale is genuinely startling once you see the figures. Roughly 50% of patients prescribed chronic medications actually stick to their treatment plans, medication nonadherence is a factor in approximately 125,000 deaths annually in the United States, and it’s linked to up to 25% of all hospitalizations, with improving adherence estimated to reduce healthcare costs by $100 to $300 billion each year. Put in perspective against other causes of death, medication non-adherence would rank as the 7th leading cause of death in the United States, exceeding deaths from influenza, pneumonia, or car accidents, and it doubles the likelihood of hospital readmission within one month after discharge.

Forgetfulness is only part of the story, and cost turns out to be a much bigger driver than most people assume. According to IQVIA data, prescription abandonment rates are under 5% when a prescription carries no out-of-pocket cost, but that rises to 45% when the cost is over $125 and to 60% when it’s more than $500. That matters for the technology angle, because a reminder can’t solve a price problem. The evidence on smart devices themselves is real but mixed: a randomized controlled trial of 61 breast cancer survivors found that a smart pill bottle reminder intervention significantly improved both medication adherence and medication self-efficacy over four weeks. That said, reviewers note the overall evidence base for adherence apps is mixed with many poor quality studies, citing small sample sizes, lack of appropriate control arms, and self-reported adherence outcomes, making firm conclusions about effectiveness difficult.

Health tracking has become the dominant reason people wear these devices at all. There are around 562.86 million smartwatch users worldwide, a 23.7% increase from 454.69 million in 2024, and over 92% of them use the device to track or improve their health and fitness. The demographic shift the article hints at is real and measurable: the 25 to 34 age bracket still accounts for the largest share of adoption at 42.6%, but seniors aged 65 and over are the fastest-growing group at 42% annually as health features improve.

It can, though the accuracy picture is more nuanced than headlines suggest, and it depends heavily on which sensor and which population. In the largest screening study of its kind, the Apple Heart Study found that 34% of individuals who received an irregular rhythm notification were later confirmed to have atrial fibrillation, with a positive predictive value of 0.84 among those notified. That means a majority of alerts in a broad screening population were not confirmed AFib, which matters for anxiety and unnecessary doctor visits. The newer ECG feature performs considerably better: a meta-analysis of 11 studies covering 4,241 participants found the Apple Watch ECG had pooled sensitivity of 94.8% and specificity of 95% for detecting atrial fibrillation compared to standard 12-lead ECG. The practical takeaway is that these devices are increasingly credible as screening tools, but a notification is a prompt to see a doctor, not a diagnosis.

There’s a concrete example that landed in Nature in late 2025, and it’s genuinely ambitious. Researchers modified the GPT architecture to model disease progression, creating a model called Delphi-2M trained on data from 0.4 million UK Biobank participants and validated on 1.9 million Danish individuals, which predicts the rates of more than 1,000 diseases based on each person’s past disease history with accuracy comparable to existing single-disease models. Its performance varies meaningfully by condition, which is worth knowing: Delphi-2M achieved the highest accuracy for conditions with predictable courses like certain cancers and myocardial infarction, while performance was lower for psychiatric disorders, pregnancy complications, and rare diseases with more variable trajectories. It also has a specific blind spot relevant to lifestyle-driven conditions: the model struggled to accurately predict Type 2 diabetes, indicating a limitation in accounting for factors not consistently captured in electronic health records.

It’s largely referring to pharmacogenomics, matching medication choice and dosing to a person’s genetics, and the numbers behind it are surprising. More than 90% of patients are thought to carry at least one genetic variant that should prompt a change in dosing or medication if certain drugs are prescribed. A long-term study reinforces just how routine this is: researchers who sequenced the genomes of 300 deceased patients with lifelong medical records found that 93% carried clinically relevant pharmacogenomic variants, and nearly 80% had been prescribed roughly three medications on average that those variants may have affected, with the FDA currently recognizing over 60 genes with variants that predict response for more than 190 drugs. The clinical problem this addresses is substantial: research has found that between 40% and 70% of patients either do not respond adequately to drug treatment or experience adverse drug reactions.

It’s a real and underappreciated one, especially for insulin. Insulin must be stored between 2 and 8°C when refrigerated and below 30°C at room temperature, with unopened vials, pens, and cartridges remaining refrigerated until their expiration date and staying usable at room temperature for up to 28 days once opened. The margin for error is narrow: even brief exposure to temperatures below 2°C or above 30°C can cause unacceptable degradation of insulin. The genuinely surprising part is that ordinary home refrigerators often fail this test without anyone noticing. A 2019 study placing Bluetooth temperature sensors next to refrigerated insulin found the temperature was out of the safe range for an average of 2 hours and 43 minutes per day, with deviations of about 3.7°C, and nearly 25% of home refrigerators dropped below 0°C, risking freezing that can render insulin completely ineffective.

This is the biggest gap the video doesn’t address, and it surprises most people. HIPAA does not cover most data generated by wearables, because that data usually stays inside consumer apps and tech platforms rather than entering the healthcare system, and HIPAA only applies once a healthcare provider, health plan, or one of their business associates actually receives, uses, or stores it. The practical consequence is that companies not bound by HIPAA can generally do whatever their terms and conditions permit with the data, including sharing and selling it, and one FTC test of 12 mobile health and fitness applications found they sent consumer data to 76 third-party companies, including device identifiers, running routes, dietary habits, and sleep patterns. This is starting to shift: the Health Information Privacy Reform Act, introduced in November 2025, aims to extend HIPAA-like protections to data generated by devices like smartwatches and smart rings, while states including California, Colorado, and Connecticut have already begun expanding their definitions of protected health information.