Study Guide to Best Understand Top-Funded DHC and Impact on High Burden Conditions: Examining the Safavi et al. (2019)
GIVEMEA Study Guide · Digital Health Innovation
Top-Funded Digital Health Companies & High-Burden Conditions
Safavi, Mathews, Bates, Dorsey & Cohen · Health Affairs 38(1):115–123 · January 2019
20Companies
156Studies Found
15%Clinical Effectiveness
28%High-Burden Studies
$2.55BTotal Equity Funding
Central Finding
Leading digital health companies have not yet demonstrated substantial impact on disease burden or cost — most studies enrolled healthy volunteers, and none of the clinical effectiveness studies measured outcomes in terms of cost or access to care.
Research Question
To what extent have the top-funded private US digital health companies studied their products in patients with high-burden, high-cost conditions — and have they measured impact on outcomes, costs, or access to care?
Company Landscape
The 20 companies spanned 12 digital health categories. Analytics (AI and big data) was most common at 25%. Biosensor companies attracted the highest funding at $706M (28% of total). Six companies (30%) were in California’s San Francisco Bay Area. Founding years ranged from 1993 to 2014; 60% founded after 2006.
Publication Findings
Of 156 total studies, 104 (67%) were PubMed-indexed. Only 15% were clinical effectiveness studies; 32% were validation studies; 53% “other.” Among the 16 clinical effectiveness studies, all measured patient outcomes — but zero measured cost or access to care. Fifty-one percent had fewer than 100 participants.
High-Burden Population Coverage
Only 27.9% of studies targeted high-burden conditions. The most studied group overall was healthy volunteers (32%), followed by ALS (14%) and multiple sclerosis (12%). Mental health was the most studied high-burden category (8 studies), yet there were zero clinical effectiveness studies for any mental health condition.
Company Type Breakdown
| Category | No. of Companies | Note | Primary End User |
|---|---|---|---|
| Analytics / AI / Big Data | 5 (25%) — most common | ~16% funding share | Hospitals, providers |
| Consumer Health Engagement | 4 (20%) | ~9% | Patients |
| Biosensors / Wearables | 3 (15%) | 28% — highest funding | Patients |
| Care Coordination | 3 (15%) | ~18% | Providers, payers |
| Population Health Mgmt. | 2 (10%) | ~10% | Patients, payers |
| Telemedicine | 2 (10%) | ~7% | Patients, employers |
| Social Networking | 2 (10%) | ~7% | Physicians |
| Other | 4 | ~5% | Mixed |
■ Business/Market ■ Research Methods ■ Policy/Regulatory ■ Technology/Clinical
Business & Market
Digital Health Company
tap to define
Digital Health Company
Entity building/selling digital health products or services — genomics, analytics, biosensors, telemedicine, mobile apps, wearables, population health tools, per FDA/Rock Health definitions.
Private Equity Funding
tap to define
Private Equity Funding
Investment capital for non-publicly traded companies. Top 20 had median funding of $67.5M vs. $5.3M for the rest — the rationale for selecting them as best-equipped to produce rigorous evidence.
Direct-to-Consumer (DTC)
tap to define
Direct-to-Consumer (DTC)
Selling to patients/consumers without clinical intermediaries. The study argues this explains why most companies targeted healthy consumers — lower regulatory hurdles and easier market access than clinical channels.
Crunchbase
tap to define
Crunchbase
Open-source database operated by TechCrunch, used to identify and rank companies by total private equity funding. The primary source for company selection and market data in this study.
Rock Health
tap to define
Rock Health
Venture fund and research org tracking digital health funding. Provided the company categorization framework (analytics, biosensors, telemedicine, etc.) adapted alongside FDA categories for this study.
Value-Based Purchasing
tap to define
Value-Based Purchasing
Payment model rewarding outcomes and cost reduction over service volume. The study argues this trend could incentivize evidence-based digital health — but notes fee-for-service still dominates.
Survivor Bias
tap to define
Survivor Bias
A limitation from selecting only currently top-funded companies. Failed high-funded companies were excluded, potentially skewing results toward more favorable appearances of the industry.
Journal Impact Factor
tap to define
Journal Impact Factor
Average annual citations per article in a journal — a proxy for scientific influence. Most common publication source (31%) had no impact factor, suggesting limited scientific visibility of digital health research.
Research Methods & Metrics
Cross-Sectional Observational
tap to define
Cross-Sectional Observational
Analyzes a defined population at a single point in time without interventions. This study’s snapshot was fixed to April 15, 2017 — after which company rankings and status changed considerably.
DALYs
tap to define
DALYs
Disability-Adjusted Life Years = YLL + Years Lived with Disability. One of three metrics defining “high-burden” conditions. Top-5 DALY conditions include depression, hypertension, obesity, heart disease.
Years of Life Lost (YLL)
tap to define
Years of Life Lost (YLL)
Deaths × standard life expectancy at age of death. One of three burden metrics. High-YLL conditions: heart disease, cancer, trauma, stroke — key digital health target areas.
USPSTF Levels of Evidence
tap to define
USPSTF Levels of Evidence
Level 1 = RCTs; Level 2 = controlled/cohort studies; Level 3 = expert opinion. Applied only to 16 clinical effectiveness studies. Results: 56% Level 1, 44% Level 2 — decent quality, but far too few studies.
Cohen’s Kappa (κ)
tap to define
Cohen’s Kappa (κ)
Interrater reliability beyond chance. Results: 0.98 (high-burden classification), 0.85 (study purpose), 0.45 (USPSTF level). Moderate kappa for evidence level reflects genuine difficulty in classifying study quality.
Clinical Effectiveness Study
tap to define
Clinical Effectiveness Study
Evaluates a product’s impact on a clinical population. Only 15% (16/104) of PubMed-indexed studies qualified. All measured outcomes; zero measured cost or access — a stark evidence gap.
Validation Study
tap to define
Validation Study
Tests whether a product measures what it claims vs. a clinical reference standard. Comprised 32% of studies — often enrolling healthy volunteers purely to confirm technical accuracy, not clinical impact.
High-Burden Condition
tap to define
High-Burden Condition
In the top 5 by cost, YLL, or DALYs (any one qualifies). Examples: heart disease, cancer, mental health, hypertension, depression, trauma, stroke. The study’s primary impact classification filter.
Policy & Regulatory
MACRA (2015)
tap to define
MACRA (2015)
Medicare Access and CHIP Reauthorization Act — strongest recent US push toward value-based care. The study uses it as a key policy lever but cautions fee-for-service still dominates and change will be gradual.
21st Century Cures Act (2016)
tap to define
21st Century Cures Act (2016)
Exempted apps intended “only for maintaining a healthy lifestyle” from FDA regulation. Makes DTC consumer markets far easier to enter without clinical evidence — a structural cause of the healthy-consumer bias in this study’s findings.
FDA Digital Health Program
tap to define
FDA Digital Health Program
FDA initiative clarifying which products require approval and at what standards, intended to support digital technologies that improve patient monitoring and clinical decision-making.
CMS Remote Patient Monitoring
tap to define
CMS Remote Patient Monitoring
Starting 2018, CMS reimbursed digital tools creating and transmitting patient-generated data to clinicians. The study proposes this model as a template for broader digital health payment reform.
FTC (Federal Trade Commission)
tap to define
FTC (Federal Trade Commission)
US regulator that polices deceptive advertising. Sued Lumosity ($2M settlement) and others for unsubstantiated health claims — underscoring the need for evidence-based digital health products even in consumer markets.
Meaningful Use / EHR Model
tap to define
Meaningful Use / EHR Model
Federal incentive program for EHR adoption with vendor certification standards. The study proposes an analogous model for digital health: certification + financial incentives for tools targeting high-burden populations.
Technology & Clinical
Biosensors / Wearables
tap to define
Biosensors / Wearables
Devices measuring physiological signals continuously. Highest-funded category ($706M). Yet most biosensor studies enrolled healthy subjects to validate technology — not to demonstrate clinical impact in sick populations.
Population Health Management
tap to define
Population Health Management
Systematic analysis of and intervention on health outcomes across a defined group. Two of the top 20 companies focused here — promising for high-burden patients but understudied in the peer-reviewed literature.
Telemedicine
tap to define
Telemedicine
Healthcare delivery via telecommunications. CMS now reimburses remote monitoring, counseling, and chronic care management — making telemedicine the clearest existing model for how digital health achieves sustainable reimbursement.
AI Diagnostics in Health
tap to define
AI Diagnostics in Health
ML/AI tools for diagnosis and clinical decision support. Most common company type (25%). The study notes AI diagnostic applications had only recently entered prospective real-world evaluation and had rarely been assessed for clinical effectiveness.
FHIR / SMART Standards
tap to define
FHIR / SMART Standards
Fast Healthcare Interoperability Resource and SMART on FHIR — federal informatics standards providing a clear technical pathway for integrating health apps into EHRs and clinical workflows.
mHealth (Mobile Health)
tap to define
mHealth (Mobile Health)
Healthcare delivery via mobile devices and apps. By 2016: 259,000 health apps existed; 296 private companies received $4.2B+ in funding — a market growing far faster than the evidence base for these tools.
Study Design at a Glance
Cross-sectional observational analysis of the 20 highest-funded private US digital health companies as of April 15, 2017. Publications identified via PubMed, Google Scholar, and company websites. Two authors independently rated all studies; a third adjudicated disagreements. Interrater reliability measured with Cohen’s kappa.
Company Selection
- Source: Crunchbase, filtered to US headquarters + FDA-aligned digital health product categories
- Ranked by: Total private equity funding through April 15, 2017
- Top 20 selected: Median funding $67.5M vs. $5.3M for rest; median founding year 2007 vs. 2012
- Excluded: Companies that had undergone an IPO (acquired products make it hard to isolate original offerings)
Publication Identification
- PubMed and Google Scholar searched by company name through April 15, 2017
- Company websites additionally searched for publication references
- Included studies both with and without company participation
- Result: 156 total; 104 PubMed-indexed (67%); 52 not indexed (33%); 19.7% year-over-year PubMed growth 2004–2016
Study Classification Framework
- Study purpose: Validation / Clinical Effectiveness / Other
- High-burden conditions: Top 5 by cost, YLL, or DALYs (defined a priori)
- Evidence level: USPSTF Levels 1–3 (applied to clinical effectiveness studies only)
- Impact measured: Outcomes / Cost / Access to care — coded for each effectiveness study
- Journal quality: Impact factor from International Scientific Institute 2015–16 ratings
Interrater Reliability (Cohen’s Kappa)
High-burden condition classification (κ = 0.98)
Very Good
Study purpose (κ = 0.85)
Very Good
USPSTF evidence level (κ = 0.45)
Moderate
Strengths
- Focused on companies best-equipped to produce rigorous research (highest funding, most experience, largest teams)
- Three independent burden metrics (cost, YLL, DALYs) to avoid single-dimension bias
- Very strong interrater agreement on the two most critical variables (κ ≥ 0.85)
- Broad publication search including non-indexed sources and company websites
- Standardized evidence hierarchy via established USPSTF framework
Limitations (Exam-Critical)
- Survivor bias: Failed high-funded companies excluded — results may overstate industry quality
- Non-random sample: Top 20 is not representative of the full digital health market
- Rapidly changing field: Snapshot from April 2017; the field was already different at publication (Jan 2019)
- Definition ambiguity: Boundary between “digital” and “non-digital” health is qualitative and shifting
- Top-5 cutoff only: Choosing only top 5 conditions per burden category excludes others that may matter
- Moderate κ for USPSTF level (0.45): Only moderate agreement on evidence quality rating
- Private sector only: Excludes academic medical centers and publicly traded companies
- Reliance on gray literature: Crunchbase and Rock Health are not peer-reviewed data sources
Analysis of the most cited and conceptually central references used to build the study’s argument.
Ref 1
Rock Health / Tecco (2016) — Year-End Funding Report
Foundational DataNon-Peer-ReviewedPrimary source for company categorization and market funding totals. Cited for the fact that 296 private digital health companies received $4.2B+ in 2016. The study both uses and critiques the landscape Rock Health documents. Key caveat: gray literature — these figures should be interpreted with that in mind.
Ref 16
Porter ME (2010) — “What is Value in Health Care?”
Conceptual AnchorProvides the foundational definition: Value = Outcomes ÷ Cost. This formula anchors the entire argument — if companies cannot demonstrate outcomes or cost reduction, they cannot claim to deliver “value.” Porter’s framework also underpins the value-based purchasing policy recommendations.
Ref 22
USPSTF Evidence Review Development (2017)
Methodology FrameworkThe authoritative source for the 3-level evidence hierarchy applied to all 16 clinical effectiveness studies. Results: 56% Level 1 (RCTs), 44% Level 2 — the clinical effectiveness studies that existed tended to be well-designed, but there were simply too few of them and none in high-burden populations.
Ref 26
Byambasuren et al. (2018) — Prescribable mHealth Apps
Corroborating EvidenceAn independent systematic review finding similarly that only a small number of mHealth products had been evaluated in RCTs. The convergence of this finding with Safavi et al.’s results across different methodologies substantially strengthens the central claim that evidence generation lags far behind product proliferation.
Ref 27
Keane & Topol (2018) — AI and Autonomous Diagnosis
Emerging EvidenceCited to contextualize AI diagnostic findings — noting these applications had only recently begun prospective real-world evaluation and few had been assessed for clinical effectiveness. Supports the position that digital health’s most hyped category also has the thinnest evidence base.
21st Cures
21st Century Cures Act (2016)
Regulatory ContextExplains a fundamental market asymmetry: wellness apps explicitly exempted from FDA regulation, making DTC consumer markets far easier to enter without clinical evidence. A key structural reason why companies target healthy consumers rather than high-burden patients.
MACRA
MACRA (2015) & CMS Value-Based Frameworks
Policy FrameworkCentral to the study’s forward-looking policy recommendations. If value-based purchasing expands, demand for proven digital tools should follow. The study cautions that fee-for-service was still projected to exceed 50% of Medicare payments at end of 2018 — so MACRA-driven change will be gradual.
Click an answer to reveal feedback. Each question locks after answering.
Question 1 of 8
Which digital health company category was most common among the top 20 funded companies?
✓ Analytics (AI/big data) was the most common category at 25%. Biosensors attracted the most funding ($706M, 28%) — a key distinction the exam tests.
Analytics (AI/big data) was most common at 25%. Biosensors had the highest funding — two different things worth keeping separate.
Question 2 of 8
What percentage of PubMed-indexed studies targeted patients with high-burden conditions or risk factors?
✓ Only 27.9% targeted high-burden populations; 72.1% did not. Most studied non-high-burden group: healthy volunteers (32%).
27.9% targeted high-burden populations. 15% is the proportion of clinical effectiveness studies — a different metric frequently confused on exams.
Question 3 of 8
Which was NOT one of the three metrics used to define “high-burden” conditions?
✓ The three metrics were cost, YLL, and DALYs — each capturing a different dimension: financial impact, premature mortality, and combined mortality + disability.
Hospital readmission rates were NOT used. The three metrics were cost, YLL, and DALYs.
Question 4 of 8
Of the 16 clinical effectiveness studies, how many measured impact in terms of cost or access to care?
✓ One of the study’s starkest findings. All 16 measured outcomes, but zero measured cost or access — directly contradicting Porter’s definition of value (Outcomes ÷ Cost).
Zero. All 16 measured patient outcomes, but none measured cost or access to care — the two metrics most relevant to value-based care policy.
Question 5 of 8
Which was the most common HIGH-BURDEN condition category studied across all publications?
✓ Mental health had 8 studies — most of any high-burden category. Paradoxically, there were zero clinical effectiveness studies for any mental health condition. Volume ≠ evidence quality.
Mental health was most studied (8 studies) but had zero clinical effectiveness studies. Being the most-studied category did not produce actionable clinical evidence.
Question 6 of 8
The 21st Century Cures Act (2016) is most relevant to this study because it:
✓ The Act exempted wellness apps from FDA regulation — making DTC consumer markets far easier to enter without evidence. This explains the healthy-consumer bias in the study’s findings.
The 21st Century Cures Act exempted wellness apps from FDA regulation — the opposite of requiring approval. This structural incentive explains why companies favored consumer markets over clinical high-burden populations.
Question 7 of 8
What Cohen’s kappa was reported for classifying study purpose?
✓ The three kappas: 0.98 (burden classification), 0.85 (study purpose), 0.45 (USPSTF level). Study purpose had very good agreement at 0.85.
Study purpose: κ = 0.85. Remember all three: 0.98 (burden condition), 0.85 (study purpose), 0.45 (USPSTF evidence level).
Question 8 of 8
The study’s primary policy recommendation was to:
✓ Two levers: (1) clarify regulatory requirements, and (2) create market incentives analogous to meaningful-use EHR incentives. The study avoids blanket restrictions — it seeks to align market incentives with patient need.
The study recommends targeted action — clarifying regulation AND building market incentives — not blanket bans. Goal: align incentives with patient need, not prohibition.
— / 8
Quiz Score
Core Thesis
Digital health holds great promise, but evidence generation has not kept pace with product proliferation. Market incentives currently favor healthy consumers over high-burden patients — and this must change through targeted policy action.
-
The Evidence Gap Is Measurable, Not Anecdotal
Only 15% of studies were clinical effectiveness studies, and zero measured cost or access to care. These are precise counts from the top-funded companies in the field — the ones best positioned to generate evidence.
-
Healthy Volunteers Are the Most-Studied “Population”
Healthy subjects represented 32% of all studies — more than any single disease category. This reflects a DTC market strategy enabled by lighter regulatory requirements, not a scientific choice about who most needs digital health tools.
-
Funding Does Not Equal Clinical Impact
Biosensors attracted the most money ($706M, 28% of total) yet produced mostly technical validation studies in non-clinical populations. High venture capital is not a proxy for clinical value and should not be treated as one in purchasing or policy decisions.
-
Regulatory Asymmetry Is a Structural Driver
The 21st Century Cures Act removed FDA oversight from wellness apps, creating a powerful incentive to target healthy consumers. Changing the evidence landscape requires changing these structural incentives — voluntary shifts alone are insufficient.
-
Value-Based Care Is the Right Lever — But Not Yet Dominant
If MACRA and accountable care expand, digital tools with proven outcomes should become commercially viable at scale. But fee-for-service still dominated at time of writing, and MACRA alone cannot transform the industry rapidly.
-
Mental Health: Most Studied, Least Proven
Mental health led all high-burden categories with 8 studies — yet had zero clinical effectiveness studies. Volume of research is not a substitute for quality. Being “studied” does not mean being proven effective.
-
AI Hype Outpaces Its Evidence Base
Analytics/AI was the most common company type (25%) but had only recently begun prospective real-world evaluation. The largest segment of the digital health market is also the segment with the thinnest clinical effectiveness evidence.
-
Two Policy Levers: Regulatory Clarity + Market Incentives
Clarify FDA regulatory requirements for products targeting sick patients, and create financial incentives — analogous to meaningful-use EHR incentives — that reward adoption of high-impact digital tools in clinical practice. Both levers are needed simultaneously.