Universal health coverage and nursing and midwifery workforce growth from 2000 to 2023: A longitudinal ecological study

Highlights

• •Growth profiles compared coverage and workforce development in 194 countries.
• •Matching endpoint years retained 96.9% of country growth profiles.
• •Coverage and workforce density were positively associated within countries.
• •Country profile labels depended on the thresholds and period examined.
• •Workforce monitoring should include employment, deployment and retention.

Abstract

Background

National service coverage and workforce density describe different aspects of health-system development. Comparing their rates of change may reveal patterns overlooked by static workforce counts.

Objective

To examine how growth in universal health coverage service coverage relates to growth in nursing and midwifery personnel density across national health systems.

Design

Longitudinal ecological study.

Settings

National health systems from 2000 to 2023.

Participants

Country-year records: growth profiles included 194 countries; adjusted models included 2290 observations from 169 countries.

Methods

We analysed World Health Organization and World Bank indicators. Separately standardised annualised growth rates defined a relative alignment score, with median thresholds distinguishing four profiles. Two-way fixed-effects models estimated associations between service coverage and log workforce density. Sensitivity analyses examined matched endpoint years, alternative growth measures and thresholds, country trends, first differences and country influence.

Results

The reference analysis identified 44 coverage–workforce gap countries, with above-median coverage growth and below-median workforce growth. Matching endpoint years preserved 188 of 194 profiles (96.9%) and retained 41 gap countries. Alternative thresholds identified 30–63 gap countries; the fixed 2010–2023 comparison produced 70.1% agreement among 87 eligible countries. The adjusted coverage coefficient was 0.0294 (95% confidence interval 0.0175–0.0413), equivalent to approximately 3.0% higher workforce density per one-point increase in the index. Coefficients remained positive with country-specific trends (0.0396; 0.0182–0.0611) and first differences (0.0248; 0.0144–0.0351).

Conclusions

Coverage and workforce density were positively associated, but relative growth varied between countries. Matching observation windows preserved most profiles, whereas thresholds and period affected membership. Joint monitoring should consider growth rates, starting levels and reporting quality before investigating workforce constraints. Incomplete reporting, measurement limitations and the ecological design preclude conclusions about staffing adequacy or causal effects.

Graphical abstract

Universal health coverage and nursing and midwifery workforce growth from 2000 to 2023: A longitudinal ecological study

The reference growth analysis classifies 44 countries in the relative gap profile; matching endpoint years for both indicators retains 41. Gap membership means above-median coverage growth and below-median nursing and midwifery density growth. Counts vary with thresholds and period. The adjusted coefficient describes association with log density; country-trend and first-difference checks remain positive. Profiles do not establish adequate staffing, and coefficients do not establish causality.

What is already known.

• •Summary box describing the contribution of nurses and midwives to essential services, documented workforce shortages and unequal distribution, and the limits of national workforce density as a measure of employment, deployment, workload or care quality.

What this paper adds.

• •Summary box describing growth profiles across 194 countries, profile sensitivity to endpoint years, thresholds and period, and a positive within-country association that does not establish causality.

1. Background

Universal health coverage requires access to needed health services without financial hardship. Its two central dimensions, service coverage and financial protection, remain distinct in global monitoring (World Health Organization and World Bank, 2025). Neither measure, however, establishes whether a health system has the workforce needed to sustain care. Workforce planning therefore requires a complementary assessment of whether the personnel available to deliver services are increasing alongside service coverage.

Nurses and midwives provide preventive, maternal and newborn, chronic disease and continuing care across community and hospital settings. Their contribution makes workforce development directly relevant to the expansion of these services. Nursing is the largest occupational group in the health workforce, and the 2025 global nursing report identifies education, employment and service delivery as central policy concerns (World Health Organization, 2025). We used a combined nursing and midwifery indicator because it captures both professions within the available data. Although the measure cannot distinguish their respective contributions or working conditions, it provides a consistent focus for examining national workforce development.

This focus also reflects the interdependence of the six health-system building blocks identified by the World Health Organization: service delivery; health workforce; information; medical products, vaccines and technologies; financing; and leadership and governance (World Health Organization, 2007). Workforce availability can constrain the benefits of investment in facilities or financing reforms. Conversely, additional personnel need appropriate skills, funded posts, equipment and deployment to contribute effectively to care. Examining coverage and workforce development together therefore connects changes in service provision with one of the resources required to sustain them.

Previous ecological research has examined the relationship between workforce availability and service coverage. The GBD 2019 Human Resources for Health Collaborators (2022) estimated workforce densities in relation to an effective-coverage target across 204 countries and territories, while Boniol et al. (2022) described international workforce stocks, distribution and projected shortages. Building on this evidence, we examine how annualised changes in a reported service coverage index compare with changes in nursing and midwifery density within a common sample of countries. Combining relative growth profiles with estimates of within-country association allows us to examine national variation around the average relationship. The analysis complements existing assessments of workforce availability without proposing a new global staffing threshold.

1.1. Objectives

The study asked whether service coverage and nursing and midwifery workforce development moved together between 2000 and 2023, and how national growth patterns varied around their average relationship. We aimed to quantify annualised growth in both indicators, describe variation using a continuous alignment score and four median-based profiles, examine annual trajectories within those profiles, and estimate the association between coverage and workforce density within countries. Sensitivity analyses assessed the influence of observation windows, growth measures, classification thresholds and model specification. Maternal mortality provided exploratory context for interpreting differences between profiles.

1.2. Conceptual framework

Health labour market theory provides a basis for interpreting these relationships. It distinguishes the education of qualified workers from their participation in the labour force and employment in funded posts; migration, retention and distribution further shape the workforce available for service delivery (Sousa et al., 2013). Additional education places may have limited immediate effect if graduates cannot enter employment. Equally, expanded financing or service entitlements may require changes in deployment as well as headcount. These mechanisms could influence the relationship between coverage and workforce growth, although they were not measured in the present dataset.

Fig. 1 locates the two measured national indicators within these possible labour-market, organisational and clinical pathways. Where service demand increases without sufficient staffing, heavier workloads may contribute to missed care and occupational strain. Associations between workload and care left undone, and between burnout and poorer safety or quality, support further investigation of these pathways (Ball et al., 2014; Li et al., 2024). Such findings inform the conceptual framework but cannot establish that countries with high relative growth scores experienced these consequences. The service coverage index itself does not directly measure service volume or demand.

Fig. 1.

Measured national indicators and proposed pathways. The upper panel identifies the service coverage index and nursing and midwifery density analysed in this study. The lower panel summarises possible labour-market, organisational and clinical pathways that were not measured. Arrows do not represent estimated causal effects.

The framework informed two analytical propositions. First, increases in the service coverage index within countries would be positively associated with nursing and midwifery density. Second, this average association could coexist with a subset of countries showing above-median coverage growth and below-median workforce growth. We examined the first proposition through panel regression and the second through descriptive growth profiles. The latter constitutes a classification exercise rather than an independent causal hypothesis test. Neither proposition defines the workforce growth required for safe or equitable care.

2. Methods

2.1. Design and analytical sample

We conducted a longitudinal ecological study of national indicators from 2000 to 2023. Ecological studies analyse groups rather than individuals; the unit of analysis here was the country-year for panel models and the country for growth profiles. This design addresses questions about national development, but it cannot establish relationships between individual workers and patients (Wakefield, 2008). The dataset comprised 4704 country-year records for 196 countries and territories, referred to collectively as “countries” throughout the paper.

Eligibility varied by analysis. Descriptive summaries included all available indicator values within the study period. Growth profiles required at least two positive observations in distinct years for each central indicator, while regressions required positive workforce density and complete data for every variable in the relevant specification. We included all eligible records and undertook no prospective sample-size calculation. The accompanying checklist adapts the Strengthening the Reporting of Observational Studies in Epidemiology principles to this aggregate longitudinal design (von Elm et al., 2007). It guides reporting completeness without certifying methodological validity.

2.2. Data sources and measures

The reported universal health coverage service coverage index, nursing and midwifery personnel density, hospital-bed density and maternal mortality ratio were obtained from the World Health Organization Global Health Observatory. Current health expenditure came from the World Bank World Development Indicators. Supplementary Table S1 lists the source codes and units. The retained files record updates in December 2025 for coverage, January 2026 for workforce density, September 2025 for hospital beds and April 2025 for maternal mortality. These dates indicate source modifications; the original download dates were not established. We restricted all archived series to the common study period of 2000–2023.

Nursing and midwifery density was expressed as personnel per 10,000 population. We retained the combined indicator as supplied, recognising that national reporting systems may differ in occupational coverage and in whether they count active, practising or registered personnel. The measure represents workforce availability at national level. It does not capture funded employment, full-time-equivalent staffing, skill mix, geographical distribution or workload (World Health Organization, n.d. a).

The service coverage index ranges from 0 to 100 and combines tracers for reproductive, maternal, newborn and child health; infectious diseases; non-communicable diseases; and service capacity and access (World Health Organization, n.d. b). We used the published index without reconstructing its components or the estimation procedures underlying individual observations. The index describes service coverage rather than the volume of services delivered, financial protection or care quality. Its service-capacity component may also overlap with broader measures of health-system development.

Current health expenditure as a percentage of gross domestic product and hospital beds per 10,000 population represented time-varying financing and infrastructure conditions. The expenditure measure records the share of national income devoted to health, rather than spending per person or the budget for nursing salaries (World Bank, n.d.). Maternal mortality, expressed as deaths per 100,000 live births, was summarised by profile using 2023 values. It provided descriptive context and was neither a regression outcome nor a basis for estimating the health effects of misalignment.

2.3. Harmonisation and missing data

We linked indicators by country code and calendar year, using numeric source fields rather than formatted display strings. Each country-year record was unique. Comparison with the five retained source extracts confirmed that all non-missing analytical values matched their source values. We neither interpolated nor carried values forward, and we did not impute additional observations. The published source series may themselves incorporate modelling by the international agencies.

Available observations entered each analysis according to its eligibility criteria. Workforce density and hospital beds were reported less completely than service coverage. Because missingness may reflect reporting capacity and health-system conditions, complete-case models may represent a selected sample. We therefore report sample sizes for each indicator and model, although the observed records cannot establish the direction or magnitude of selection bias.

2.4. Annualised growth and relative profiles

For each country, we calculated annualised growth separately for the two central indicators using their earliest and latest positive observations within the study period. For an indicator X observed in years t₀ and t₁, the annualised percentage growth rate was:

$$g_i=\left[{\left(X_{i,t_1}/X_{i,t_0}\right)}^{1/\left(t_1-t_0\right)}-1\right]\times 100$$

In the reference analysis, each indicator retained its own observation window. Annualisation adjusts for differences in elapsed time, but it does not align calendar periods or eliminate sensitivity to endpoint errors and low initial values. The country-profile dataset records the endpoint values, years and number of available observations. Because workforce growth is calculated from density, it reflects changes in both personnel numbers and the population denominator.

We standardised each annualised growth variable across eligible countries using its own mean and sample standard deviation. The alignment score for country i was then calculated as standardised coverage growth minus standardised workforce growth:

$$A_i=\left(g_{U,i}-g{¯}_U\right)/s_U-\left(g_{W,i}-g{¯}_W\right)/s_W$$

Here, U and W denote the coverage index and workforce density, respectively. A positive score indicates that coverage growth occupies a higher standardised position in its own distribution than workforce growth occupies in its distribution. It does not necessarily indicate a higher raw percentage growth rate for coverage. Similarly, a zero score denotes equal standardised positions, rather than equal percentage growth, proportional capacity or adequate staffing. The score thus provides a comparison within the analytical sample.

We defined four descriptive profiles using the unrounded median growth rates. Aligned expansion comprised countries with growth at or above the median for both indicators. The coverage–workforce gap profile combined coverage growth at or above its median with workforce growth below its median; workforce-led growth represented the reverse combination. Structural stagnation comprised countries below both medians. These labels refer to relative growth: structural stagnation need not imply declining services or poor performance, and aligned expansion does not establish sufficient staffing. Classifications close to either threshold warrant particular caution.

For each profile, we report mean growth rates, mean alignment scores and descriptive 95% confidence intervals around the latter. Intervals were calculated from the country-level standard deviation using the t distribution with profile size minus one degrees of freedom. They reflect between-country variation under an assumption of independence and exclude uncertainty in the source data and classification. As the profiles and alignment score are derived from the same growth variables, differences between profile means are expected by construction and provide no independent validation of the typology.

2.5. Annual trajectories

To examine development over time within these profiles, we standardised each indicator using a fixed mean and sample standard deviation pooled across its available country-year observations for the 194 eligible countries in 2000–2023. We then plotted annual means, giving each reporting country equal weight and retaining its reference profile throughout. Standardisation was common to all years and profiles. Annual denominators are included in the reproducibility files because the countries contributing observations may change over time. Separation between the curves represents relative indicator positions and cannot be interpreted as a numerical staffing shortfall.

2.6. Panel regression

We used two-way fixed-effects linear models to estimate the association between service coverage and the natural logarithm of nursing and midwifery density. Country effects accounted for time-invariant national characteristics, while year effects accounted for common annual influences. The models therefore estimated associations from changes within countries. The adjusted specification was:

$$ln\left(W_{it}\right)=\beta U_{it}+{\gamma }^{\prime }{X}_{it}+{\alpha }_i+{\lambda }_t+{\epsilon }_{it}$$

In this specification, W denotes workforce density and U the coverage index for country i in year t. The vector X contains expenditure and hospital-bed density; α and λ denote country and year effects, respectively, and ε is the residual. Model 1 included coverage and both sets of fixed effects. Model 2 added expenditure and beds. We then refitted Model 1 to the complete-case sample used in Model 2 to distinguish covariate adjustment from changes in sample composition. Model 3 replaced contemporaneous coverage with its value in the preceding calendar year while retaining current covariates. Lagged values were matched to the preceding calendar year, rather than the previous non-missing observation.

Standard errors were clustered by country, with a finite-sample correction based on the numbers of countries, observations and estimated parameters, including fixed effects. Confidence intervals and two-sided p-values used the t distribution with the number of countries minus one degrees of freedom (Cameron and Miller, 2015). Coefficients are unstandardised semi-elasticities: a one-point increase in the coverage index corresponds to a percentage difference in density of 100 × [exp(β) − 1]. We assessed statistical significance at 5%, placing emphasis on estimates and confidence intervals.

The one-year lag examined temporal ordering without representing the full duration of education and employment pathways. Income-group interactions could not be estimated because the retained file lacked a verified classification. We defined model fit as the proportion of residual variation explained after removing country and year effects from both the outcome and predictors. This two-way within-model measure differs from overall or country-only coefficients of determination.

The reproducibility files contain the data, source checks, country-level calculations, annual plotting denominators, analysis protocol and executable Python code using NumPy, pandas and SciPy. They also record software versions and numerical outputs. Coefficients obtained by partialling out fixed effects were checked against direct dummy-variable fits, including for the country-trend and first-difference specifications described below.

2.7. Additional sensitivity analyses

We specified a protocol for additional sensitivity analyses during revision, after the reference results were known but before examining the additional results. To assess the influence of observation windows, we recalculated growth between the earliest and latest years with positive observations for both indicators in each country. Medians and standardised scores were recalculated within each eligible sample. We also applied the reference numerical cut-offs to distinguish changes in growth from threshold recalibration. Restricting harmonised windows to at least 10 and 15 years assessed sensitivity to short observation periods. A separate fixed 2010–2023 comparison required observed values for both indicators at both endpoints, without interpolation.

To assess the growth measure, we fitted log-linear trends to all paired years for countries with at least five observations spanning at least 10 years. Annual growth was calculated as 100 × [exp(b) − 1], where b is the slope from regressing the logarithm of each indicator on year. For classification sensitivity, we independently combined the 40th, 50th and 60th percentile thresholds for each indicator, giving nine definitions for each of the reference and harmonised growth datasets. We compared country transitions, percentage agreement and Spearman correlations between scores. Recalibrating reference profiles within each eligible subset also helped distinguish the influence of period or growth definition from sample composition.

We assessed score sensitivity by winsorising growth at the 1st and 99th percentiles before standardisation and by taking the difference between empirical percentile ranks. Omitting each country in turn assessed its influence on median-based profiles and the adjusted regression. A balanced trajectory comparison required paired observations in every year from 2000 to 2023, retaining the reference profiles and full-sample standardisation. To assess model specification, we added country-specific linear trends to the adjusted model and fitted a first-difference model with year effects, relating annual changes in log density to changes in coverage, expenditure and beds. First differences required complete records exactly one calendar year apart. Both specifications used country-clustered inference, with parameter ranks accounting for the applicable effects and trends. The supplement reports all specified checks.

2.8. Ethics and involvement

The study used publicly available, aggregated country-level secondary data. No identifiable individual data were accessed and no participants were recruited; ethical approval and informed consent were therefore not required. Nurses, midwives and patients were not directly involved in the design. Research incorporating their experiences remains necessary to interpret these national patterns in relation to practice.

3. Results

3.1. Analytical sample and indicator coverage

Of the 4704 country-year records, 4680 contained service coverage values, representing 195 countries, and 2981 contained workforce density values, representing 194 countries. Hospital-bed data covered 2958 observations from 193 countries, and expenditure data covered 4529 observations from 192 countries (Table 1). The growth analysis included 194 countries. Puerto Rico and the occupied Palestinian territory were ineligible because the retained extracts contained no workforce-density observations for them. The adjusted regression included 2290 complete observations from 169 countries; the lagged model included 2225 observations from the same 169 countries.

Table 1.

Indicator-specific analytical coverage and descriptive statistics, 2000–2023.

Indicator and unit	Observations / countries	Mean (standard deviation)	Median (interquartile range)	Minimum–maximum
Service coverage index (0–100)	4680 / 195	62.78 (18.49)	69.00 (49.00–77.00)	13.00–92.00
Nursing and midwifery personnel per 10,000 population	2981 / 194	48.24 (37.89)	41.23 (16.39–68.06)	0.88–208.30
Maternal deaths per 100,000 live births	4680 / 195	173.84 (254.49)	57.77 (13.77–231.33)	0.11–1662.00
Hospital beds per 10,000 population	2958 / 193	34.48 (26.28)	27.73 (14.90–47.72)	1.54–222.70
Current health expenditure (% of gross domestic product)	4529 / 192	6.23 (2.87)	5.65 (4.13–7.98)	1.22–27.09

Values summarise available country-year observations. Countries contribute different numbers of records. Growth profiles and regressions use separate eligibility rules.

Mean service coverage was 62.78 on the 0–100 scale. Nursing and midwifery density averaged 48.24 per 10,000 population, with a median of 41.23 and a range of 0.88–208.30. These pooled statistics summarise available country-year observations rather than equally weighted national averages. Median annualised growth was 0.74% for coverage and 1.48% for workforce density. Observation windows differed between the two indicators at one or both endpoints in 126 of the 194 countries.

3.2. Relative growth profiles

Applying the median thresholds yielded 53 countries in aligned expansion, 44 in the coverage–workforce gap, 44 in workforce-led growth and 53 in structural stagnation (Table 2; Fig. 2). Mean annualised growth in the gap profile was 1.78% for coverage and −0.73% for workforce density, compared with 0.42% and 3.09%, respectively, in workforce-led growth. These averages summarise variation within each profile and do not describe every country. The relatively balanced profile sizes also reflect the use of sample medians for classification.

Table 2.

Annualised growth profiles and descriptive health-system context.

Relative growth profile	Countries	Mean annual coverage growth (%)	Mean annual workforce growth (%)	Mean alignment score (95% confidence interval)	2023 maternal mortality mean / median
Aligned expansion	53	2.25	5.47	-0.10 (-0.43, 0.23)	199.2 / 142.0
Coverage–workforce gap	44	1.78	-0.73	1.32 (0.99, 1.64)	208.0 / 150.7
Workforce-led growth	44	0.42	3.09	-1.00 (-1.15, -0.85)	28.4 / 14.4
Structural stagnation	53	0.39	0.16	-0.16 (-0.26, -0.06)	33.8 / 18.7

Profiles use the unrounded medians: coverage growth 0.7375676% and workforce growth 1.4829384% per year. Alignment is the difference between separately standardised annualised growth rates, using sample standard deviations. Maternal mortality is per 100,000 live births; it is available for 53, 44, 44 and 52 countries, respectively. All mortality values are from 2023. Profiles and score intervals are descriptive and do not establish adequate staffing or causal effects.

Fig. 2.

Annualised coverage and workforce growth in 194 countries. Each point represents one country. Dashed lines mark the sample medians, displayed as 0.74% and 1.48% per year; classification uses unrounded values. The four profiles describe relative growth, not absolute workforce adequacy. Each indicator uses its own earliest and latest positive observations.

Mean alignment scores reflected these contrasts in component growth rates. The gap profile had a mean of 1.32 (95% confidence interval 0.99–1.64), compared with −1.00 (−1.15 to −0.85) in workforce-led growth. The corresponding means were −0.10 (−0.43 to 0.23) for aligned expansion and −0.16 (−0.26 to −0.06) for structural stagnation (Fig. 3). As the scores and profiles share the same underlying variables, these differences are descriptive rather than independent evidence validating the classification.

Fig. 3.

Mean relative alignment scores by growth profile. Points indicate country-level means and horizontal bars indicate descriptive 95% confidence intervals based on the t distribution. Zero means equal positions in the separately standardised growth distributions; it does not indicate equal raw growth rates or adequate staffing. Profile contrasts are expected from their construction and are not an independent validation test.

Initial indicator levels also differed substantially between profiles. Median earliest available coverage values were 29.0 in aligned expansion, 38.5 in the gap profile, 72.5 in workforce-led growth and 70.0 in structural stagnation. Median earliest workforce densities were 7.59, 17.90, 35.18 and 56.22 per 10,000 population, respectively (Supplementary Table S2). Although the initial observations did not necessarily refer to the same year, these contrasts show that relatively slow growth can coexist with higher coverage and workforce density.

Maternal mortality in 2023 varied alongside these differences in initial conditions. Mean ratios were 199.2 in aligned expansion, 208.0 in the gap profile, 28.4 in workforce-led growth and 33.8 in structural stagnation (Table 2). Values were available for every country except Niue, which belonged to structural stagnation.

This unadjusted comparison does not separate growth alignment from baseline conditions and other national characteristics. The complementary within-country regression estimates for log nursing and midwifery density are presented in Table 3 and discussed in Section 3.4.

Table 3.

Two-way fixed-effects models of log nursing and midwifery density.

Predictor or model information	Model 1 All eligible records	Model 1 Common sample	Model 2 Adjusted	Model 3 One-year lag
Service coverage index	0.0298 (0.0203, 0.0392)	0.0338 (0.0210, 0.0466)	0.0294 (0.0175, 0.0413)	0.0265 (0.0144, 0.0386)
Health expenditure (% of gross domestic product)	—	—	0.0035 (-0.0176, 0.0246)	0.0043 (-0.0169, 0.0254)
Hospital beds per 10,000 population	—	—	0.0080 (0.0043, 0.0117)	0.0081 (0.0042, 0.0120)
Observations	2981	2290	2290	2225
Countries	194	169	169	169
Two-way within-model coefficient of determination	0.108	0.151	0.217	0.193
Country and year fixed effects	Yes	Yes	Yes	Yes
Standard-error clustering	Country	Country	Country	Country

Entries are unstandardised coefficients with country-clustered 95% confidence intervals. Model 3 uses the previous calendar year’s coverage index and current covariates. Inference uses a finite-sample cluster correction and country-cluster degrees of freedom. The fit statistic is 1 minus residual sum of squares divided by the sum of squared outcome residuals after projection on both country and year effects; it measures explanatory power beyond those fixed effects.

3.3. Country examples and annual trajectories

Table 4 illustrates country-level variation through the ten highest and five lowest alignment scores, together with their component growth rates and observation windows. Rwanda had the highest score, with annual coverage growth of 5.30% during 2000–2023 and workforce growth of 0.77% during 2008–2023. The United Republic of Tanzania also had a high positive score, yet belonged to aligned expansion because its unrounded workforce growth lay just above the median. The example shows how a continuous score and a median-based profile describe different aspects of the same data. Short observation windows, including South Sudan’s 2018–2022 workforce window, require particular caution.

Table 4.

Countries at the extremes of the relative growth-alignment distribution.

Country and profile	Coverage years	Coverage growth (% per year)	Workforce years	Workforce growth (% per year)	Alignment score
Rwanda Coverage–workforce gap	2000–2023	5.30	2008–2023	0.77	3.95
Myanmar Coverage–workforce gap	2000–2023	2.13	2005–2022	-7.49	3.62
Djibouti Coverage–workforce gap	2000–2023	1.83	2002–2022	-6.15	2.96
Mali Coverage–workforce gap	2000–2023	2.36	2004–2023	-4.45	2.92
Guinea Coverage–workforce gap	2000–2023	3.61	2000–2022	-0.31	2.80
Madagascar Coverage–workforce gap	2000–2023	3.49	2002–2023	-0.63	2.78
Cambodia Coverage–workforce gap	2000–2023	3.85	2000–2019	0.64	2.73
United Republic of Tanzania Aligned expansion	2000–2023	3.97	2002–2022	1.48	2.59
Malawi Coverage–workforce gap	2000–2023	3.24	2004–2023	-0.42	2.50
Cameroon Coverage–workforce gap	2000–2023	1.78	2004–2022	-4.70	2.49
Chile Aligned expansion	2000–2023	1.19	2010–2023	11.94	-2.93
South Sudan Aligned expansion	2000–2023	2.95	2018–2022	17.06	-2.90
Antigua and Barbuda Workforce-led growth	2000–2023	0.28	2007–2019	9.11	-2.90
Guatemala Workforce-led growth	2000–2023	0.15	2009–2023	7.05	-2.40
Paraguay Aligned expansion	2000–2023	1.06	2002–2022	8.19	-1.94

The ten highest scores and five lowest scores are shown. Scores compare standardised positions in separate growth distributions, not workforce adequacy. Category membership uses unrounded growth medians. The United Republic of Tanzania has annual workforce growth 1.483154%, just above the 1.482938% median; both round to 1.48%. Observation windows differ, and short windows can amplify endpoint sensitivity. Harmonised-window and profile-sensitivity results are reported in Section 3.5 and the supplement.

The annual means in Fig. 4 place these endpoint comparisons in their temporal context. Trajectories varied over time, with no uniform pattern of divergence throughout the period. Their separation also reflected the indicators’ different global distributions and changes in the countries reporting data. The plots therefore describe annual patterns without formally testing sustained decoupling or quantifying a shortfall in delivery capacity.

Fig. 4.

Annual standardised indicator means within fixed growth profiles, 2000–2023. Solid lines show the service coverage index and dashed lines show nursing and midwifery density. Each indicator uses a fixed pooled mean and sample standard deviation across its available country-year values in the 194-country growth sample. Annual means weight reporting countries equally; their denominators may vary. Supplementary Fig. S2 shows a selected balanced-panel comparison. The plots do not measure a staffing shortfall or establish sustained causal decoupling.

3.4. Within-country associations

Model 1 yielded a coverage coefficient of 0.0298 (95% confidence interval 0.0203–0.0392). Fitting the same specification to the adjusted-model sample gave 0.0338 (0.0210–0.0466). After adjustment for expenditure and hospital beds, the coefficient was 0.0294 (0.0175–0.0413; p < 0.001), equivalent to approximately 3.0% higher workforce density per one-point increase in the coverage index within a country (Table 3). The adjusted model had a two-way within-model coefficient of determination of 0.217.

Hospital-bed density was positively associated with workforce density in the adjusted model, with a coefficient of 0.0080 (95% confidence interval 0.0043–0.0117). The expenditure coefficient was 0.0035 (−0.0176 to 0.0246), with no clear evidence of an independent association in this specification. Using lagged coverage produced a coefficient of 0.0265 (0.0144–0.0386; p < 0.001), equivalent to approximately 2.7% higher density per one-point increase in the previous year’s index. These conditional associations do not estimate the effects of changes in coverage policy.

3.5. Sensitivity of growth profiles and associations

Matching the two indicators’ endpoint years within each country retained all 194 countries, of which 188 kept their reference profile (96.9% agreement). The gap profile included 41 of the original 44 countries; Jamaica, Norway and the Russian Federation moved to structural stagnation. Within the harmonised gap profile, mean annual growth was 1.96% for coverage and −0.76% for workforce density. Alignment scores were strongly correlated with the reference rankings (Spearman correlation 0.975). Applying the reference numerical cut-offs instead yielded 44 gap countries, indicating the contribution of threshold recalibration (Supplementary Tables S3–S4).

Profile agreement remained high when harmonised windows were restricted to at least 10 or 15 years: 96.3% among 190 countries and 94.3% among 176 countries, respectively. Log-linear trends gave 86.3% agreement among 182 countries, with 40 in the gap profile. Agreement was lower in the fixed 2010–2023 comparison, at 70.1% among 87 countries. This comparison identified 18 gap countries, including nine of the 15 original gap countries eligible for that analysis. Recalibrating reference medians within the same 87 countries increased agreement to 75.9%, indicating that sample composition contributed to the difference alongside the period or growth measure.

Classification was also sensitive to the choice of thresholds. Across the nine combinations, gap membership ranged from 30 to 63 countries using reference growth and from 30 to 62 using harmonised growth; 29 countries belonged to the gap profile under all 18 combinations (Supplementary Fig. S1). Winsorised scores remained highly correlated with the unmodified rankings (0.997 and 0.998), while percentile-rank differences produced correlations of 0.929 and 0.927. Omitting one country at a time changed at most three remaining profiles in the reference analysis and two in the harmonised analysis. The fully balanced trajectory panel contained 44 countries, including only four original gap countries, limiting this comparison to a selected sample (Supplementary Fig. S2).

The positive regression association persisted across the additional specifications. With country-specific linear trends, the adjusted coverage coefficient was 0.0396 (95% confidence interval 0.0182–0.0611; 2290 observations, 169 countries). In first differences, it was 0.0248 (0.0144–0.0351; 1944 annual changes, 142 countries). Across 169 models omitting one country at a time, adjusted coefficients ranged from 0.0272 to 0.0311, and every confidence interval remained above zero. These checks support the stability of the association under the specifications examined, without establishing its causal direction (Supplementary Table S6).

4. Discussion

4.1. Within-country associations and national growth patterns

Service coverage and nursing and midwifery density were positively associated within countries, although national growth patterns varied around this average relationship. A subset of countries combined comparatively rapid coverage growth with slower workforce growth. Matching endpoint years preserved 188 of the 194 reference profiles, and the positive association persisted in models with country-specific trends and first differences. The analyses address complementary questions: regression estimates conditional associations within countries, while the profiles compare relative national growth. Together, they describe variation in development without identifying a common causal mechanism or a staffing requirement.

Examining this variation extends assessments based on workforce density at a single point in time. The GBD 2019 Human Resources for Health Collaborators (2022) related workforce availability to an effective-coverage target; our analysis compares changes in a reported service coverage index with changes in a combined nursing and midwifery indicator. Differences in coverage measures, occupational scope, modelling and periods preclude direct comparison of thresholds or effect sizes. Both approaches nevertheless support assessing workforce levels alongside their development over time.

National context remains essential to that assessment. Boniol et al. (2022) showed that global workforce totals conceal substantial distributional inequalities. Similarly, the meaning of growth depends on a country’s starting position: increasing density from a low base may leave a substantial shortage, whereas slow growth at high coverage may reflect a ceiling effect. The higher initial values in the two profiles with slower coverage growth make this distinction particularly relevant here. Profile names therefore describe comparative growth and should not be used to rank health systems or diagnose policy failure.

The sensitivity findings further qualify the interpretation of profile membership. Matching endpoint years changed six classifications and retained 41 of the original 44 gap countries. However, alternative thresholds identified between 30 and 63 gap countries, and the fixed 2010–2023 analysis retained nine of the 15 original gap countries represented in that subset. The broad pattern of relative growth was therefore more stable than the precise country list. Although 29 countries remained in the gap profile across all 18 threshold and window combinations, this consistency establishes stability under those rules, not clinical understaffing or heightened risk. Country assessment still requires continuous growth rates, initial levels and local evidence.

4.2. Health-system and labour-market interpretations

The health-system building-blocks framework offers one explanation for the positive average association: workforce, service delivery, infrastructure and financing develop within interdependent systems (World Health Organization, 2007). The positive association between hospital beds and workforce density is consistent with such joint development, although it does not identify the underlying mechanism. Coverage may respond to workforce changes, and both may respond to investment, governance or population needs. Country and year effects cannot remove all these time-varying influences.

Where workforce growth is comparatively slow, the health labour market framework directs attention to possible constraints in education, funded employment, geographical deployment, migration and retention. The present analysis cannot determine which constraint applies. Instead, the profiles identify patterns that can be investigated using labour-market information before a response is chosen. This approach recognises that expanding training alone may not increase the workforce available for care. It follows the distinction between workforce supply and funded demand set out by Sousa et al. (2013) and in the health labour market analysis guidebook (World Health Organization, 2021).

The expenditure result illustrates the importance of choosing measures that correspond to these mechanisms. Expenditure as a share of gross domestic product reveals neither the allocation to nursing and midwifery posts nor the reliability of salary payments or graduates’ entry into employment. The share can also change when national income changes. The imprecise coefficient therefore indicates that this aggregate measure had no clearly distinguishable association with workforce density after adjustment; it does not establish that health financing is unimportant.

4.3. Implications for care and the workforce

For nursing and midwifery practice, the relevant concern is whether rising service expectations are accompanied by sufficient staff available for deployment. Facility-level evidence associates higher workload with care left undone (Ball et al., 2014), providing a reason to examine staffing, time for monitoring and continuity of care where national patterns warrant closer investigation. Translating those patterns into practice requires direct organisational evidence, because the growth score cannot quantify nurse-to-patient ratios or missed care.

The same distinction applies to occupational strain and clinical outcomes. A recent systematic review associated nurse burnout with poorer reported safety and quality, including more missed care, medication errors and healthcare-associated infections (Li et al., 2024). That evidence is heterogeneous and largely observational. Burnout and clinical outcomes were not measured in our study, so these associations inform possible pathways for further research rather than establish consequences of profile membership. Assessing retention likewise requires direct information on employment exits and working conditions, which national density alone cannot provide.

Staffing decisions may also have financial implications, although their direction and magnitude depend on context. In a prospective hospital-panel study in Queensland, improved staffing following ratio legislation was associated with better patient outcomes; estimated savings from fewer readmissions and shorter stays exceeded the additional staffing costs (McHugh et al., 2021). This finding supports considering financial and clinical consequences together in that setting. It cannot provide a transferable estimate of the cost of national coverage–workforce misalignment.

The maternal mortality comparison demonstrates why national outcomes also require cautious interpretation. Mortality was higher in both profiles with faster relative coverage growth, including aligned expansion, rather than following a simple gradient in alignment scores. Baseline disadvantage, differences in access and care quality, and wider social and health-system conditions may explain the pattern. Further analysis would need appropriate adjustment and explicit temporal ordering before relating mortality to growth alignment. The present comparison establishes neither harm from coverage expansion nor protection from membership of a slower-growth profile.

4.4. Implications for planning and monitoring

For national planning, the findings suggest a staged assessment. An initial review should establish indicator definitions, reporting completeness, starting levels and calendar windows. Examination of continuous growth rates and sensitivity to alternative profile definitions can then identify patterns requiring closer attention. National and subnational information on education output, funded posts, vacancies, entry into employment, deployment and retention is needed to investigate the underlying constraints. This sequence links comparative monitoring to the workforce investment priorities in the 2025 global nursing report (World Health Organization, 2025), while avoiding the use of profile membership as an automatic staffing target.

The appropriate response will depend on what that investigation reveals. Rising graduate numbers alongside persistent vacancies may warrant examination of recruitment, financing and deployment. Growth in registered personnel without evidence of active practice may instead indicate a need for better employment data. Service expansion plans should specify the settings and services requiring staff and how posts will be funded and sustained. These implications draw on the conceptual frameworks and wider evidence; the effects of such interventions were not estimated here.

Adding this temporal perspective can strengthen workforce monitoring while preserving service coverage and financial protection as distinct dimensions of universal health coverage. The alignment score supplements those measures and should not replace them. Facility and subnational research is needed to connect national patterns with workload, continuity, care quality and equitable access, and to establish whether similar growth profiles arise from different mechanisms across settings.

4.5. Strengths and limitations

The study draws on more than two decades of country-year data, reports sample eligibility explicitly and provides reproducible analyses. Matching observation windows addresses an important source of incomparability. Longer windows, fitted trends and alternative thresholds show how profile membership responds to analytical choices, while country trends, first differences and country-omission checks assess the stability of the regression association. These analyses strengthen the interpretation but leave several limitations.

Incomplete workforce and hospital-bed reporting may reflect differences in national reporting capacity, limiting the representativeness of the complete-case sample. Definitions and reporting practices may also change over time. We retained the international agencies’ published values and could not recover the uncertainty or methodological changes underlying each observation. The regression and profile confidence intervals therefore exclude uncertainty in the source measurements.

Growth estimates remain sensitive to measurement changes, low initial values and short observation periods. Harmonisation aligns the two indicators within each country but does not impose a common period across countries. The fixed-period analysis addresses that distinction, although it includes only 87 countries and produces greater changes in profile membership. Similarly, the balanced trajectory panel includes only 44 countries, with four in the gap profile, and cannot establish representativeness. Alternative thresholds and scores assess sensitivity to specified rules; they do not validate natural groupings or quantify uncertainty in the source data and classifications.

The indicators also impose limits on interpretation. National density cannot describe active employment, working hours, occupational mix or distribution within countries. The bounded coverage index measures neither service volume nor clinical demand, financial protection or care quality. Ceiling effects and its service-capacity component complicate interpretation of the association with workforce density. We did not reconstruct the index without capacity tracers, so potential overlap between the measures remains unresolved.

Finally, time-varying confounding, joint determination and reverse causation remain possible despite the use of fixed effects, lags, country trends and first differences. Differencing can amplify measurement error and changes both the eligible sample and the timescale of association. Country-clustered inference does not fully account for dependence between countries. The additional checks were specified during revision after the reference results were known and are not preregistered confirmatory tests. Maternal mortality provides unadjusted context only. These limitations confine the conclusions to relative national growth patterns and conditional associations.

5. Conclusions

Service coverage and nursing and midwifery density were positively associated within countries across models. Matching endpoint years preserved most profiles, but membership remained sensitive to thresholds and period. Joint monitoring should consider growth rates, starting levels, matched windows and sensitivity findings when investigating education, employment, deployment and retention. These comparisons inform workforce assessment without establishing staffing adequacy or causal effects.

Funding

This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors.

Declaration of generative artificial intelligence and assisted technologies

During preparation and revision, the authors used ChatGPT-assisted computational tools to support code development, statistical checking, table and figure preparation, source identification, language editing and manuscript organisation. The authors are responsible for reviewing the source evidence and all outputs, including the analysis, references and interpretation, and for the final content of the submitted manuscript.

Ethics approval and consent

This study analysed publicly available aggregate country-level data. No identifiable individual data were accessed, and no individuals were recruited. Ethical approval, consent to participate and consent for publication were not required.

CRediT authorship contribution statement

Zawadi Ally Nkulikwa: Writing – original draft, Visualization, Methodology, Investigation, Formal analysis, Data curation, Conceptualization, Writing – review & editing. Racheal Mukoya Masibo: Writing – review & editing, Validation.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

The source indicators are publicly available from the World Health Organization Global Health Observatory and World Bank World Development Indicators. Supplementary reproducibility files provide the analytical country-year data, retained source extracts, source-check results, country profiles, annual plotting denominators and executable analysis and figure code. Source terms continue to apply to the underlying data. No claim is made that these materials have been deposited in an external public repository.