Category: Population Health

New Model Predicts 900 Diseases From Real Records

Posted on by Jason Barry

Last week brought a potentially significant step forward for early diagnosis in the form of a new AI model that can predict a patient’s next diagnosis across nearly 900 diseases using just real-world medical records.

Meet DT-Transformer. Researchers at Harvard, Brigham and Women’s, and the Broad Institute unveiled the GPT-style foundation model in a new arXiv preprint.

  • DT-Transformer reads a patient’s medical history as a sequence and predicts which disease will show up next, and when it might come knocking.

The real headline is the training set. DT-Transformer was trained on 57.1M structured EHR entries from 1.7M patients across MGB’s 11 hospitals and 200 clinics (2000 to 2024) – the messy reality of U.S. clinical care rather than a polished research data set.

The numbers were impressive. A few standouts:

  • DT-Transformer achieved a median AUC of 0.871 across 896 disease categories, with AUC over 0.5 for every condition.
  • The model crushed an age- and sex-based baseline by +0.214 AUC (0.871 vs. 0.657), beating it on 96% of diseases.
  • All of that was accomplished with a featherweight 2.2M parameter model that’s small enough to run just about anywhere (by comparison, Claude Fable 5 has about 6 trillion parameters).

The real test told a humbler story. When the team ran a true prospective test forecasting new diagnoses DT-Transformer had never seen, median AUC slipped to 0.713.

  • That still beats the baseline on 80% of diseases, but that gap between the retrospective flex and the prospective reality is fairly significant.  
  • A 0.871 headline number and a 0.713 crystal ball aren’t the same product, and the second one is the one that patients would actually have to deal with.

One other highlight worth mentioning: including every repeated diagnosis worsened model performance, “drowning out the signal” rather than improving predictions – another reminder that more data isn’t always better. Better data is.

The Takeaway

Population-scale risk forecasting that runs on a model smaller than most phone apps is a real milestone, and training on routine records instead of a spotless biobank is exactly the kind of thing that could get models like DT-Transformer in front of actual patients – assuming the data holds up to peer-review.

Employers Feeling the Healthcare Pressure

Posted on by Jason Barry

Employers are among digital health’s most important customer bases, and the Business Group on Health’s annual survey is one of the best ways to get a pulse on their healthcare priorities.

Some stats that caught the eye:

  • Employers expect healthcare costs to spike 8% in 2025 – the highest jump in a decade.
  • Pharmacy has risen from 21% to 27% of employer healthcare spend from 2021 to 2023.
  • Cancer, MSK, and cardiovascular conditions are the top three cost drivers.
  • 80% of employers feel pressure for new mental health, obesity, and cancer treatments.

The above chart shows the accelerating climb in employer healthcare costs since 2019, with most of the 28% increase looking like it was absorbed by employer contributions to premiums. 

  • Cancer was the most commonly cited cost driver in 2024, and 72% of employers noted a higher prevalence of the disease among their workers and their families.
  • While some of that can be attributed to delayed care and screenings during the pandemic, it’s also perpetuated by an “alarming” rise of diagnoses in younger people.

GLP-1s, gene therapies, and specialty drugs were another major culprit behind the cost pressure, and over half of employers specifically called out the impact of GLP-1s.

  • Nearly all employers cover GLP-1s for diabetes, compared to 67% for obesity and 34% for cardiology (although it appears that cardiology’s share could double by 2027).
  • GLP-1s are undoubtedly promising medications, yet 96% of employers worry about their long-term cost implications, and 52% would strongly consider reducing their coverage to better control healthcare expenses.

To reign in costs, employers are actively reevaluating their vendor partnerships, and plan to use upcoming RFPs to negotiate better terms and end relationships with underperformers.

  • Six in 10 employers are looking to cut back on the number of vendors they work with, and underutilized solutions will be the first ones on the chopping block.
  • Vendor management is at the center of employers’ cost containment strategies, and they’re aiming to integrate benefits to eliminate expenses and simplify the member experience.

The Takeaway

Employers are clearly feeling the pressure to get costs under control, but the areas where they need the most help – obesity care, behavioral health, MSK – are also brimming with digital health startups promising that their solution is the answer. That begs the question: if these tools are actually working, why are costs only continuing to climb faster?

Are Preventable Hospitalizations A Flawed Measure?

Posted on by Jason Barry

Preventable hospitalizations are one of healthcare’s most widely adopted quality measures, which is exactly why they ended up in the crosshairs of a new opinion piece in Health Affairs.

Rates of preventable hospital admissions were first developed as an indicator of access to timely ambulatory care over 30 years ago, yet they’re now used to judge the performance of hospitals, health plans, and even individual providers (MSSP is an easy example).

  • These measures are typically based on inpatient admissions or ED visits for conditions like hypertension or asthma, where hospitalizations can potentially be avoided if patients have access to effective ambulatory care.

The authors argue that the metamorphosis of preventable hospitalizations from an access measure to a quality indicator was a serious misstep, and incorporate nearly a dozen studies to help make their case.

  • One of the most rigorous studies was from the Agency for Healthcare Research and Quality, which found that differences in socioeconomic status explain “a substantial part – perhaps most” of the variation in preventable admissions for many conditions, and recommended against using the metric as a standalone way to assess provider quality.

Those findings prompted the director of the AHRQ to publish an editorial outlining the “persistent challenge of avoidable hospitalizations,” which referenced a VA trial suggesting that high-quality ambulatory care might result in more, not fewer, admissions.

  • The VA analyzed 1,400 veterans hospitalized with potentially avoidable admissions for congestive heart failure, diabetes, and COPD. Veterans that received intensive primary care were hospitalized more frequently than those in the usual care group, but were also more satisfied with their care despite the additional hospitalizations.

Other evidence suggests that using preventable hospitalizations as a measure of provider quality can actually decrease care quality by discouraging necessary admissions.

  • This study linked reduced readmissions to increased mortality for heart failure patients, unexpected results that might be explained by the fact that “incentives to avoid readmissions may lead to potentially inappropriate management of higher-risk patients… in the outpatient rather than inpatient setting.”

The Takeaway

Whether or not preventable hospitalizations are a flawed quality measure, this article is a solid reminder that “strong beliefs should be loosely held,” and that widely adopted views are often the most important ones to stress-test.

Closing the Women’s Health Gap

Posted on by Jason Barry

Despite living longer than men, women spend a quarter of their lives in worse health, prompting the World Economic Forum and the McKinsey Health Institute to publish an in-depth report to unpack the underlying factors driving the disparity.

The report explores the root causes of the women’s health gap, which span far beyond sexual and reproductive health even though the area is a go-to oversimplification for many gender differences. (Chart: Contributing Factors to Women’s Health Burden)

  • Just 5% of women’s health burden is related to gender-specific conditions (maternal and gynecological)
  • 47% stems from conditions that affect women either disproportionately (depression, auto-immune disease) or differently (AFib, colon cancer)
  • 43% comes from conditions that don’t affect women disproportionately or differently (ischemic heart disease, tuberculosis, etc.)

The large disparity resulting from conditions that impact everyone more-or-less equally can be attributed to several systemic issues, which the report does a great job outlining alongside possible solutions.

  • Better clinical trial design is needed to ensure equitable representation for conditions that affect women differently, such as incorporating male vs. female disease prevalence mix and using sex-specific thresholds for biomarkers.
  • Accurately assessing the prevalence of conditions such as endometriosis and menopause is needed to improve the notoriously underestimated metrics, which leads to underinvestment due to misunderstood market potential. 
  • Enhancing access to gender-specific care is critical to improving outcomes, which might include health systems implementing new guidelines (e.g, sex-specific cutoffs for biomarkers, discharge checklists) to guide decision making and minimize biases.

In typical McKinsey fashion, the report devotes significant real estate to the economic burden of the women’s health gap, which if closed could boost global GDP by $1T due to improved workforce participation. It’s also worth noting that 10 conditions contribute over half of the economic burden. (Chart: Economic Burden by Condition)

The Takeaway

Tackling the women’s health gap is essential for far better reasons than boosting GDP, but regardless of the justification, progress depends on addressing the issues outlined in this report. Glaring research gaps, disparities in care delivery, and underinvestment have led to massive disparities in women’s health, but they’ve also created a huge opportunity for those that can help to solve them.

Social Needs Impact on Diabetes Management

Posted on by Jason Barry

New research published in JAMA Network Open found that health-related social needs (HRSNs) have a major impact on both quality and utilization outcomes for people with type 2 diabetes.

Using self-reported data from a national sample of 21,528 Medicare Advantage beneficiaries with T2D, researchers found that the majority (56.9%) reported at least one HRSN, and that each one significantly affected outcomes.

Here’s a look at the odds ratio impact of HRSNs on diabetes medication adherence (proportion of days covered over 80%), statin adherence (PDC over 80%), and having controlled HbA1c. Not too surprising to see that financial strain had the largest negative impact on HbA1c control.

  • Food insecurity – Diabetes (0.93), Statin (1.02), HbA1c (1.02)
  • Financial strain – Diabetes (0.91), Statin (0.91), HbA1c (0.83)
  • Loneliness – Diabetes (0.85), Statin (0.79), HbA1c (0.96)
  • Unreliable transport – Diabetes (0.80), Statin (0.80), HbA1c (0.94)
  • Housing insecurity – Diabetes (0.78), Statin (0.96), HbA1c (0.92)

Each of the five HRSNs also influenced hospital utilization, and it was interesting that food insecurity was identified as the factor with the largest impact on acute care usage. Here are the changes in utilization per 1,000 MA enrollees for avoidable hospitalizations, ED visits, inpatient encounters, and 30-day readmissions.

  • Food insecurity – Hospitalization (17.1), ED Visit (84.6), IE (30.4), Readmission (8.2)
  • Financial strain – Hospitalization (4.6), ED Visit (40), IE (6.8), Readmission (2.3)
  • Loneliness – Hospitalization (3.9), ED Visit (173), IE (-6.3), Readmission (-4.5)
  • Unreliable transport – Hospitalization (13.5), ED Visit (244.6), IE (41.8), Readmission (1)
  • Housing insecurity – Hospitalization (6.3), ED Visit (55.4), IE (16.1), Readmission (10.2)

The Takeaway

Type 2 diabetes affects more than a quarter of people above age 65 while costing the US health system over $200B each year, and this study underscores the significant role that HRSNs play in managing the disease. Much of the existing work around HRSNs relies on area-level social needs measures because patient-level data is hard to come by, but these findings suggest that collecting patient data is worth the effort to move the needle in diabetes care.

Misdiagnoses Linked to 370k Deaths Annually

Posted on by Jason Barry

A new study in The BMJ attracted a lot of attention last week after finding that the full impact of misdiagnoses in the U.S. is likely being seriously underestimated by the medical community.

The researchers estimate that 371k people die every year following a misdiagnosis, and 424k are permanently disabled – meaning nearly 800k people suffer “serious harm” annually. 

The lead author of the paper, Johns Hopkins Center for Diagnostic Excellence Director David Newman-Toker, said that “the number of diagnostic errors that happen out there in the U.S. each year is probably somewhere on the order of magnitude of 50 to 100 million.”

  • While these misdiagnoses don’t usually result in serious harm (most people aren’t seeing a doctor for a life-threatening condition), the study found that just 15 diseases account for about half of all misdiagnoses. 

Five diseases on their own – stroke, sepsis, pneumonia, venous thromboembolism, and lung cancer – caused almost 40% of the total “serious harm” incidents due to misdiagnosis.

  • That equates to 150k prevented deaths or disabilities every year if misdiagnoses could be cut in half for just those conditions.

The authors’ solution? Take a cue from the heart attack model. Although heart attacks frequently involve vague symptoms like general chest pain, they have less than a 2% chance of being misdiagnosed (vs. 18% for stroke). 

  • The study attributes that success to a decade of concentrated efforts, which started with recognizing that misdiagnosis was a problem and culminated with heavy research investments, new guidelines, and tighter requirements around performance monitoring.
  • “You end up ultimately with a system of care that focuses on not missing heart attacks. It’s the model for what we could be doing.”

The Takeaway

Although 800k annual incidents is an alarming total, the study concludes on the optimistic note that there’s still less than a 0.1% chance of serious harm related to misdiagnosis after a healthcare visit. That said, there’s clearly more to be done around improving the diagnosis of diseases that have severe consequences when missed, and this research does a good job highlighting the areas that should be a priority.

PwC: Healthcare Costs Projected to Rise 7%

Posted on by Jason Barry

PwC’s annual medical cost report is out, and as always, it’s a lot to dig through. The full report breaks down each of the factors underpinning medical inflation, but the headline takeaway was that healthcare costs are projected to rise 7% in 2024.

That’s up from 5.5% in 2022 and 6% in 2023, with PwC attributing the acceleration to two primary changes.

Inflator 1: In an inflationary environment like the one we’re currently seeing, providers are pushed to seek significant rate increases from payors. That effect is only amplified by workforce shortages and the return of patient demand.

  • PwC expects inflation to cause employers to focus on high performing networks, centers of excellence that target high-cost claims (particularly cancer and orthopedic cases), and health plans that steer patients to lower cost providers.

Inflator 2: Pharmacy expenses are skyrocketing, with the median annual cost of newly approved drugs reaching $222k in 2022. Combined with the introduction of new cell and gene therapies, we look poised for persistent double digit pharmacy inflation.

  • The demand of GLP-1 medications is another strong contributor, and if the FDA decides that these drugs are appropriate for weight loss, their popularity is just getting started.

PwC called out a couple deflationary forces, most notably biosimilar drugs entering the market at less than half the cost of their reference products. In 2023, the launch of adalimumab biosimilars to Humira was “a new milestone” in the U.S.

  • The shift to lower cost care sites also made an impact, with the report highlighting increased demand for outpatient surgeries, home-based services, and virtual care. PwC recommends that providers work with plans to establish models that share in the gains.

Cost Neutral but Key to Watch:

  • Continued efforts to manage total cost of care (the growth of value-based care)
  • The COVID hangover causing provider unit cost increases and pharmacy trends
  • Behavioral health utilization increasing 
  • Health equity’s impact on population health
  • CMS Health Plan and Hospital Price Transparency Rule
  • Medicaid redetermination 

The Takeaway

Although healthcare’s multi-year contracts dampen some of inflation’s immediate impact, PwC’s prediction of a 7% cost increase suggests that we’re now past any buffer. Industry bellwethers like UnitedHealth and Humana have already begun signaling that pent-up-demand is adding even more pressure to the equation, and PwC thinks that cost mitigation strategies should now be a top priority for payors and providers alike.

Medicare Advantage’s Favorable Selection Problem

Posted on by Jason Barry

Medicare Advantage plans could be on track to reach over $75B in overpayments this year – nearly 3x prior estimates – causing researchers at the USC Schaeffer Center for Health Policy and Economics to issue a pressing call for policy reform.

The USC study found that traditional fee-for-service Medicare beneficiaries with lower-than-average expenditures are significantly more likely to switch to Medicare Advantage plans. Favorable selection at its finest. 

  • For context, CMS sets MA rates based on the county-level expenditures of those in traditional FFS Medicare, and they’re intended for beneficiaries with average expenditures – not systematically below average.
  • As a result, risk-adjusted expenditures for the 16.9M new MA beneficiaries who made the switch from traditional Medicare between 2006 and 2019 were substantially below average, causing large overpayments due to the favorable selection effect.

This pattern of favorable selection more than doubles the $27B (6%) overpayment estimate from MedPAC for 2023, which primarily reflected “coding intensity” ($23B) and Star Rating (quality) bonuses, but didn’t include an adjustment for selection bias.

  • The researchers estimate that favorable selection alone could cause overpayments to the tune of 14.4%, which would surpass $75B when combined with MedPAC’s estimate of other factors.

The authors propose two potential strategies for improving the accuracy of MA rates:

  • Reform the current approach of linking MA rates to average expenditures of traditional Medicare beneficiaries by including measures to reduce the impact of aggressive coding and mandating new data reporting requirements to improve comparability.
  • Abandon the current approach and institute competitive bidding by MA plans to let market forces determine rates with the aim of capturing efficiency gains for taxpayers instead of increasing revenue for MA plans.

The Takeaway

Medicare Advantage enrollment has been skyrocketing over the past decade, and over half of all eligible beneficiaries are now enrolled in a private plan. As traditional Medicare enrollment continues to decline, basing MA rates on FFS expenditures will only grow increasingly problematic, and this study does a great job underscoring the need for some serious reform.

Lightbeam Acquires Jvion to Address SDOH

Posted on by Jason Barry

Population health management extends far beyond the moments of care delivery, which is why Lightbeam Health Solutions is acquiring Jvion to move from “predictive to prescriptive” intelligence, while bringing new health equity offerings to millions of patients.

Lightbeam helps payors and providers manage risk by generating patient cohorts for over 42M lives to provide proactive insights that ensure care is delivered when it can have the largest impact on a patient’s trajectory.

  • Jvion incorporates clinical, socioeconomic, and behavioral data to identify hidden patient risk factors across various diseases, then recommends appropriate clinical interventions taking into account these SDOH influences.
  • The acquisition will integrate Jvion’s AI-enabled prescriptive analytics and social determinants of health solutions with Lightbeam’s existing population health management services, allowing Lightbeam to incorporate more risk factors into its patient cohorts.  

Jvion’s SDOH solutions are well-aligned with Lightbeam’s other offerings, building an impactful health equity component into the “Deviceless RPM” services added through last year’s acquisition of CareSignal.

  • CareSignal’s Deviceless RPM service uses a combination of automated text messages and IVR calls to gather self-reported patient data, enabling value-based organizations to sustainably scale care teams without sacrificing engagement.
  • With Jvion and CareSignal both integrated into its portfolio, Lightbeam now provides a complimentary suite of services that optimizes risk profiles, suggests interventions based on SDoH factors, and continuously monitors patient progress.

The Takeaway

Integrating Jvion’s prescriptive analytics into Lightbeam’s core population health offering improves the cohort creation at the center of the service while allowing more SDoH data to feed into its insights. The combination of these health equity enhancements and CareSignal’s remote patient monitoring capabilities position Lightbeam as a highly scalable solution for end-to-end population health management, which seems like a solid value proposition at a time when labor shortages are the single biggest concern for most providers.

Truveta Launches Deidentified Clinical Data Platform

Posted on by Jason Barry

Healthcare data analytics startup Truveta announced the launch of its deidentified clinical data platform designed to provide insights on rare medical conditions and COVID-19.

Truveta was formed earlier this year by 14 health systems with a mission of “saving lives with data.” The announcement revealed that it has raised $195m to develop its platform and expand its partner base.

  • The “Truveta Platform” promises real-time answers to public health questions by aggregating partners’ deidentified patient data and communicating it in an interactive dashboard. Data inputs include all EHR data, physician notes, images, and genomics, which can then be studied based on demographics, comorbidities, and vaccine manufacturers.
  • The addition of three new health system members (Ochsner Health, Saint Luke’s, UnityPoint) pushes Truveta’s total partner count to 20, representing over 16% of US patient care from clinical sites in 42 states.
  • Preliminary insights shared in the press release found that Moderna recipients experience the most adverse events and J&J recipients have the most hospitalizations. Truveta also found that people with high-risk conditions like cancer or HIV are no more likely than the general population to have a breakthrough case, which the authors believe could be a result of risk averse behaviors.

Industry Impact

Truveta is aiming to expand its member base as quickly as possible, taking the “moral imperative” route by calling for new health systems to join its platform to improve care during the pandemic.

Although Truveta states that its data is “licensed for ethical medical research, not to target advertising to patients or physicians,” the company isn’t registered as a charity, creating tension between its mission and the path to revenue.

That said, the Truveta Platform has the potential to have a positive impact on public health by making fresh insights available from existing data, and the new funding provides significant resources to find the balance between successful health and business outcomes.

Get the top digital health stories right in your inbox

This content is exclusive to subscribers

Log in or join by entering your email below.

Completely free. Every Monday and Thursday.

DHW New Phone Image