More mobile connections and faster internet access do not automatically translate into better healthcare. A peer-reviewed analysis spanning 166 countries found that national digital connectivity was not associated with faster improvements in a broad health service coverage index when researchers examined three-year changes or changes between World Health Organization reporting years. In a separate main model of coverage levels, connectivity was associated with lower, rather than higher, coverage. That apparently counterintuitive result needs careful interpretation: the study did not test whether introducing telemedicine, electronic health records or digital clinics harms healthcare. It tested whether general communications infrastructure, by itself, tracks progress towards universal health coverage.
Published on 9 October 2026 in Frontiers in Public Health, the research by Guoliang Lu and Xinyun Tang also found a more consistent concern. Countries with a larger share of healthcare costs paid directly by households tended to have lower service coverage, even after accounting for numerous economic and institutional characteristics. Together, the results sharpen a policy distinction: access to digital infrastructure is one thing; financing, organising and delivering affordable medical services is another.
What universal health coverage actually measures
Universal health coverage, or UHC, has two connected goals. People should be able to obtain needed health services, and they should be able to do so without financial hardship. The first goal includes services such as reproductive and child healthcare, infectious disease treatment and prevention, and care for noncommunicable diseases. The second asks whether obtaining that care exposes families to unaffordable bills. Improving one does not guarantee improvement in the other.
The study’s main outcome was the WHO UHC service coverage index, a composite indicator on a scale from 0 to 100. It is not a percentage of citizens with health insurance, nor a direct count of successful medical consultations. Across the researchers’ main sample, the average index value was 63.11, with a standard deviation of 18.58. Values ranged from 14 to 92. The team also examined separate measures of financial hardship and particular healthcare outcomes because a single composite can conceal important differences.
Digital tools may help schedule appointments, share patient records, manage medicines, support remote consultations and communicate public health advice. But an internet subscription does not tell researchers whether a clinic has trained staff, whether records can move between hospitals, whether medicines are available, or whether patients can afford treatment. That difference between enabling infrastructure and actual implementation sits at the centre of the study.
How the researchers compared 166 countries
Lu and Tang assembled a panel using multiple international datasets covering 2000 to 2024. Their principal statistical analysis included 3,522 country-year observations from 166 countries between 2001 and 2023. A country-year represents one country’s data for one year, so the sample does not mean 3,522 independent countries or 3,522 individuals. The authors used a two-way fixed-effects model that accounted for persistent differences between countries and for shocks shared across particular years. Standard errors were clustered by country.
Their digital connectivity measure combined three indicators: the proportion of people using the internet, mobile cellular subscriptions per 100 residents, and fixed broadband subscriptions per 100 residents. The researchers standardised the measures and averaged at least two available components. Importantly, this index describes general telecommunications access. It does not measure whether doctors actually use digital systems or whether those systems work well.
Other model variables included the share of current health expenditure financed by government, the share paid out of pocket by households, government effectiveness, gross domestic product per person, urbanisation, population age structure, drinking-water and sanitation access, and several broad indicators of the online political information environment. Exposures were generally lagged by one year, so the model compared an earlier characteristic with subsequent coverage. Lagging can improve temporal ordering, but it does not make an observational association causal.
The researchers did more than estimate one equation. They examined country-specific trends, region-by-year comparisons, first differences, different lag structures, WHO reporting-year intervals, three-year changes, income-group patterns, health workforce and hospital capacity, and exploratory measures of vaccination and mortality. They also compared connectivity with the Global Digital Health Monitor’s 2023 measures of digital-health maturity. These additional analyses are essential because a strong result in one statistical specification can become much weaker when the question or sample changes.
The headline finding changes with the statistical question
In the principal model of UHC coverage levels, a one-standard-deviation increase in the digital connectivity composite was associated with a 2.98-point lower service coverage index. The reported 95% confidence interval was from −4.05 to −1.91 points, with p below 0.001. The result is statistically distinguishable from zero in that model, but it is not evidence that installing broadband reduces healthcare quality. Countries may expand connectivity while their health systems face other constraints, and the model cannot eliminate all changing sources of confounding.
More importantly for the question of whether connectivity predicts improvement, the association was not clear in the longer-change analyses. For a three-year change in UHC, the estimate was −0.17 index points, with a 95% confidence interval from −0.66 to 0.32. For annualised changes between WHO reference years, the estimate was also −0.17, with a confidence interval from −0.38 to 0.04. Both intervals include zero. In plain language, the study could not establish that countries becoming more connected subsequently improved their broad health coverage faster over those windows.
Other model choices further tempered the original negative coefficient. In a within-between decomposition, the within-country estimate was −0.93 points, with a confidence interval from −1.94 to 0.09. Adding country-specific trends produced an estimate of −0.77 points, with a confidence interval from −1.32 to −0.22. A region-by-year specification yielded −1.03, with an interval from −2.10 to 0.05. These estimates use different statistical comparisons and scales, so they should not be averaged together or interpreted as one stable causal effect.
One measurement detail helps explain the caution. The published UHC index series contains modelled, interpolated and integer-valued annual estimates. The authors found that 40.6% of adjacent country-year index changes were zero. An outcome that changes slowly, or is smoothed between reporting years, may be poorly suited to detecting the early effects of new digital services. Conversely, a negative level association can reflect factors that the model has not fully captured. Neither problem justifies concluding that digital healthcare is ineffective.
Household medical bills showed a clearer association
The out-of-pocket expenditure share told a different story. A one-standard-deviation higher share of current healthcare expenditure paid directly by households was associated with a 1.38-point lower UHC service coverage index, with a 95% confidence interval from −2.39 to −0.36 and p = 0.008. In the main sample, the average household out-of-pocket share was approximately 35.0% of current health expenditure, although national circumstances varied widely.
The government health expenditure share, by contrast, had an estimated coefficient of −0.71 points, with a confidence interval from −1.62 to 0.20 and p = 0.128. That estimate did not clearly differ from zero in the principal model. This does not establish that government spending is unimportant. Government and household financing shares are parts of the same total, and the authors report a substantial negative correlation between them. A shift in one share changes the composition of the whole. How money is raised, pooled, allocated and spent may matter more than a single percentage in isolation.
An alternative measure, the logarithm of total current health expenditure per person, was positively associated with coverage: 2.28 index points, with a confidence interval from 0.33 to 4.24. But different expenditure measures capture different questions, and none of these observational coefficients can predict what would happen if a particular government increased its health budget tomorrow.
Internet infrastructure is not the same as digital health readiness
The 2023 Global Digital Health Monitor comparison makes the infrastructure distinction tangible. General connectivity correlated strongly with overall digital-health maturity (Spearman rho = 0.73), infrastructure (0.73), leadership and governance (0.69), and legislation and policy (0.69). Yet the relationship was far weaker for strategy and investment (0.16), standards and interoperability (0.20), workforce readiness (0.32), and actual services and applications (0.28). The overall-maturity comparison included 151 countries, while some individual domains had fewer observations.
These figures are correlations, not implementation scores or percentages. They indicate that a country with extensive telecommunications networks is not necessarily equally advanced at training clinicians, connecting incompatible records, purchasing reliable software or bringing digital services into everyday patient care. This helps explain why connectivity can be a necessary enabler without producing measurable service gains on its own. A related Research Today analysis of national genomics initiatives similarly distinguishes the presence of advanced technology from the governance structures needed to use it responsibly.
Vaccination and newborn outcomes did not follow one pattern
The authors also explored specific health outcomes and applied a false-discovery-rate procedure to reduce the risk of treating chance results as discoveries across many tests. Of 80 secondary main-effect tests, four met their exploratory threshold of q at or below 0.10. Greater connectivity was associated with 4.62 percentage points lower coverage for the third dose of diphtheria, tetanus and pertussis vaccine, and 3.74 points lower coverage for the first measles-containing vaccine dose. These estimates were observational and should not be read as evidence that internet use discourages vaccination.
At the same time, higher connectivity was associated with approximately 12% lower neonatal mortality in a log-scale model. Higher government health expenditure share was associated with 12.86 percentage points higher second-dose measles vaccine coverage. These mixed directions matter. A digital or institutional characteristic can coincide with different patterns across vaccination, newborn survival and an aggregate service index. The findings do not support a simple story that connectivity is uniformly beneficial or uniformly harmful.
Exploratory interaction analyses suggested that government effectiveness and the broader information environment might modify some immunisation relationships. For example, the connectivity by government-effectiveness interaction for second-dose measles coverage was 6.84, while the connectivity by adverse-information-environment interaction was −5.98 in their respective model units. However, comparable interactions were not established for the overall UHC index. The political information measures were broad national indicators, not direct measures of whether individual parents saw vaccine misinformation. Mechanisms remain hypotheses rather than demonstrated causes.
Why the study cannot settle the digital health debate
There are several limitations. First, countries are the units of analysis. National averages conceal inequalities within countries, including differences between cities and rural communities and between people with different incomes, disabilities or levels of digital literacy. Second, the researchers measured access to telecommunications infrastructure, not telemedicine adoption, electronic medical record use, software quality or the experiences of patients and health workers. Third, health-system organisation, purchasing arrangements, interoperability and implementation quality were incompletely observed.
Fourth, country and year fixed effects and one-year lags do not remove every potential confounder. Changes in health policy, reporting practices, economic shocks or disease burdens could influence both connectivity and measured coverage. Fifth, some financial protection data were sparse. The analysis of households spending more than 10% of their resources on healthcare, for example, included only 775 observations from 149 countries. A nominal association between government financing and that measure did not survive adjustment for multiple testing.
Finally, the changing sample itself mattered. When the researchers restricted analyses to countries and years with detailed physician, nurse and hospital-bed data, the connectivity coefficient was much smaller than in the full sample, even before adding those capacity measures. That suggests missing data and sample composition can materially alter the estimated relationship. Results from different model specifications should not be treated as interchangeable.
What governments and healthcare leaders can take from it
The study points towards practical questions for policymakers. Does a connectivity investment reach clinics that lack reliable communications? Can clinicians use the resulting systems without extra administrative burden? Can a patient move between facilities without repeating tests because records cannot be shared? Are medicines and staff available once a digital consultation identifies a need? And can lower-income households afford the services to which digital tools direct them? Those are implementation questions that a national mobile subscription rate cannot answer.
For countries considering digital public health programmes, evaluation should therefore track service use, referral completion, patient waiting times, treatment continuity, affordability and who benefits, rather than counting devices and network connections alone. The research also argues for stronger longitudinal evidence that measures digital-health deployment directly, ideally alongside changes in health outcomes at regional or facility level.
The conclusion is narrower, but more useful, than a claim that technology has failed healthcare. Across 166 countries, general connectivity did not reliably predict faster improvements in broad UHC service coverage in the study’s change models. Higher household out-of-pocket financing was associated with lower coverage. The lesson is that a connected health system still has to be a functioning, financed and equitable health system.
Source Information
Original study: Health financing, digital connectivity, and universal health coverage: a country-year panel study.
Authors: Guoliang Lu and Xinyun Tang.
Journal: Frontiers in Public Health, Health Economics section, volume 14.
Publication date: 9 October 2026.
Research type: Peer-reviewed observational multi-source country-year panel analysis with fixed-effects models and sensitivity analyses.
Study coverage: Principal model of 3,522 country-year observations from 166 countries, 2001 to 2023.
DOI: 10.3389/fpubh.2026.1928583.
