Impacts of COVID-19 mobility-restricting policies and perceived COVID-19 risks on household unmet medical needs in Nigeria: a difference-in-differences analysis

Abstract

Background

The coronavirus disease (COVID-19) pandemic and mobility-restricting policies (MRPs) have created substantial barriers to healthcare access globally. This study quantified the causal impact of government-imposed MRPs and perceived COVID-19 risk on household unmet medical needs in Nigeria, examining the differential effects for vulnerable populations.

Methods

We conducted difference-in-differences analyses using nationally representative household panel data from the pre-pandemic period (January-February 2019, n = 1,596 households) and the early pandemic period (April-August 2020). Unmet needs were measured as self-reported forgone medical care at the household level. We employed two approaches to disentangle voluntary from policy-driven restrictions: comparing states with total versus partial lockdown, and comparing states with below- versus above-median mobility reductions (Google Mobility data). We estimated COVID-19 risk effects across four state-level case burden categories during and after the lockdown.

Results

Households in total lockdown states experienced 13.0% points (95%CI: 3.0–23.0) (doubling baseline rate) higher unmet needs compared to households in partial lockdown states. This converges with mobility approach showing 15.0% points (95%CI: 6.0–25.0) higher unmet needs among individuals with below-median mobility. During lockdown, the COVID-19 case burden had no significant differential impact. However, during the post-lockdown period (June-August 2020), households in Lagos (the epicentre, with more than 10,000 cases) experienced 24.0% points (95%CI: 2.0–46.0) higher unmet needs compared to low-risk states. Persons with disabilities faced disproportionate barriers during lockdown, with a 40.0% point (95% CI: 6.0–74.0) higher rate in high-risk states and a 74.0% point (95%CI: 22.0-126.0) higher rate in the epicentre. No differential impacts were observed for poverty or chronic disease status.

Conclusion

Considering the detrimental effects of unmet medical needs, this study suggests that policymakers should evaluate the risks of COVID-19 in relation to the implementation of MRPs to protect households and vulnerable groups during future pandemics in Nigeria.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12913-026-14086-0.

Introduction

The World Health Organization (WHO) estimated that there were 2.74 times higher excess deaths than the 5.42 million deaths attributed to the coronavirus disease (COVID-19) worldwide between 2020 and 2021 [1]. Several similar reports of excess mortality above that caused by COVID-19 have been reported in many countries [2, 3]. The Sub-Saharan Africa (SSA) region experienced the highest global excess mortality between 2020 and 2021, and Nigeria had one of the highest in the region, with an estimated excess mortality of 53·66 times that from COVID-19 [3]. Like previous pandemics in the region, unmet medical needs for other diseases were mainly responsible for these excess deaths [4, 5].

An ‘unmet need’ occurs when an individual or household cannot access needed healthcare [6]. Millions of people worldwide have been severely affected by unmet needs during the COVID-19 pandemic, increasing the risk of morbidity and mortality from reasons other than the virus [7, 8]. People with non-communicable diseases (PWNCDs) and people with disabilities (PWD) are particularly at risk of these detrimental outcomes from unmet needs [9, 10]. During the pandemic, people were unable to access healthcare due to a perceived fear of contracting infection and government-imposed mobility-restricting policies (MRPs) [7]. These also had incidental effects on the supply of medical personnel and equipment [11], leading to unmet needs (Supplementary Fig. 1).

Following the first case of COVID-19 recorded in Nigeria, the Federal Government of Nigeria, on 30 March 2020, announced a range of MRPs that included movement restrictions, bans on religious and social gatherings, closure of schools, offices and businesses, and dusk-to-dawn curfew in major cities [12]. Many states also adopted similar measures with some variability, and by early May 2020, these MRPs, otherwise known as ‘lockdown’, had been announced in many states [13]. These measures, one of the most stringent in SSA, lasted 36 days and were gradually eased out from 5 May to 30 June [13]. By this time, Nigeria had also recorded Africa’s second-highest number of confirmed COVID-19 cases [14]. Although there is evidence that MRPs helped control COVID-19 infection and undoubtedly prevented related morbidities and mortalities [15]. There is an ongoing debate on whether these benefits outweigh the resulting adverse effects from unmet needs [16]. Evaluating these trade-offs is crucial in SSA, including Nigeria, where high excess mortality rates and a low COVID-19 fatality ratio exist.

An accurate estimation of the impact of MRPs on unmet needs requires disentangling the confounding effect of voluntary self-imposed mobility restrictions (SMRs) occasioned by the perceived risks of contracting COVID-19 [17]. These SMRs were observed in some countries long before the implementation of MRPs [18]. The impacts of the MRPs and SMRs from COVID-19 on unmet needs among Nigerian households during the pandemic have not been quantified. This study aims to quantify the causal impact of government-imposed MRPs and perceived COVID-19 infection risks on household unmet medical needs in Nigeria during the first wave of the pandemic (April-August 2020), and to examine whether these impacts differed for vulnerable populations, including households with disabilities, non-communicable diseases, and those in poverty. However, it has been methodologically challenging to separate the impact of MRPs and SMRs [17]. We addressed this challenge by estimating two different difference-in-differences (DID) equations using state-level lockdown data (which measured MRPs) and Google Mobility data (which measured overall restriction). For robustness, we also separately estimated the impact of the number of COVID-19 cases on unmet medical needs using a DID design. A DID design is a quasi-experimental method that compares changes in COVID-related outcomes before and after the introduction of a given policy or intervention in a given area, to changes in the same outcomes where no such policy or intervention was implemented [17]. It is superior to other methods used in studying the effects of the pandemic on healthcare utilisation [19–21] because it employs a control group to make causal inferences [17].

Method

Data sources

We used data from the first phase of the monthly national longitudinal phone surveys (NLPS) conducted by the Nigeria National Bureau of Statistics (NBS) with support from the World Bank (WB). The NLPS sample was drawn from a pre-existing face-to-face General Household Survey Panel (GHSP). GHSP, collected between January and February 2019, was the fourth wave of a nationally representative panel survey that began in 2010. It comprises 5,000 households randomly sampled across a primary sampling unit of 500 enumeration areas (EAs). Detailed information regarding GHSP can be found elsewhere [18].

The NLPS collected socioeconomic (income, employment, food security) and health access data through phone interviews during the pandemic to determine the effects of the pandemic, including accompanying restriction measures, on households. The data for the first phase of NLPS were collected from April 2020 to April 2021. A sample of 3,000 households was randomly selected for interviews. The response rate from the sample selected from 4,440 households whose members’ phone numbers were obtained during the original GHSP-W4 was more than 90% (Supplementary Table 1).

We used the first four rounds of NLPS, which included information on health access (Supplementary Table 1). These rounds spanned most of the first wave of the pandemic (April-August 2020). We also used the face-to-face GHSP-W4 data, which provided an excellent source of pre-pandemic information for our study.

Outcome measures

Self-reported ‘unmet medical needs’ in the survey, which captures the presence of a medical need that was not met due to non-consultation or inability to access care. It assigns a value of 1 if anyone in a household had a medical need but did not access care for it, and 0 if all household members who needed medical care accessed it. In the GHSP, it was assessed by non-consultation for needed care in the 4 weeks before the interview, while in the NLPS, it was assessed by the inability to access needed care starting from mid-March 2020. It captures foregone healthcare due to any access barrier, regardless of the underlying reason.

Explanatory variables

The explanatory variables were three indicators of mobility restrictions, capturing government-mandated MRPs and voluntary SMRs resulting from a perceived risk of COVID-19 infection. Government-imposed MPRs were captured via two explanatory variables: [1] state-level lockdown type (total vs. partial, binary-coded based on policy reports); [2] Google Mobility data (percentage change in visits to grocery/pharmacy and retail/recreation sites, aggregated to classify states as reduced vs. increased mobility depending on if it is below or above the median cumulative value). In addition, voluntary SMRs due to perceived COVID-19 risks were proxied by state-level cumulative confirmed cases, categorised into low (< 100), medium (100–1000), high (> 1000-10,000), and epicentre (> 10,000) based on the Nigeria Centre for Disease Control (NCDC) data. The details are discussed below.

The impact of mandatory mobility restrictions policies on households’ unmet needs during the first wave of the pandemic

Using two complementary measures of MPRS, we employed a DID model to estimate the impact of mobility restrictions on unmet needs. Fitting two separate regressions for each measure provided robustness to our analysis while also indirectly isolating the impact of voluntary SMRs from that of government-imposed MRPs. We hypothesised that if SMRs were minimal, results from both approaches would converge.

Partial-total lockdown approach

We leveraged spatial variation in the stringency of government-announced MRPs (partial or total lockdown) across Nigerian states to construct our treatment variable. Based on federal and state government announcements collected from official policy documents and verified through local media reports (see Supplementary Table 2), we classified states into two categories: (1) Total lockdown (in Lagos, Ogun, Kano, FCT, and 14 others where complete 24-hour restriction on non-essential movement, mandatory closure of all non-essential businesses and markets, and prohibition of inter-state and intra-state travel) (Fig. 1a) (2) Partial lockdown (in the remaining 19 states where time-restricted curfews (typically 8pm-6am), closure of schools and religious gathering spaces, restriction of large gatherings, but essential businesses remained open with reduced hours and capacity limits) (Fig. 1a). Therefore, our explanatory or treatment variable is binary, taking the value of 1 for households residing in total lockdown states and 0 for those in partial lockdown states (control group).

Fig. 1.

States that implemented partial and total lockdown (a) and states with reduced and increased mobility to ‘grocery and pharmacy’ and ‘retail and recreation’ relative to the median values for Nigeria during April/May 2020 lockdown using the Google Mobility data (b). Notes: States were categorised into partial and total lockdowns based on government announcements. States were categorised into those with reduced mobility and increased mobility compared with the median mobility for ‘Grocery and Pharmacy’ and ‘Retail and Recreation’ categories in Google Mobility Data, compared with their baseline values (as of January 3– February 6, 2020) during the lockdown in Nigeria

Google mobility report approach

In this approach, we used aggregated, anonymised mobility data from Google Community Mobility Reports (Google Limited Liability Company (LLC) 2020) to construct an objective measure of actual mobility changes, which reflects the combined effects of both government policies and voluntary behavioural changes, as discussed below. We chose Google mobility data due to its high penetration rates and representativeness [22].

Google Mobility data measured the percentage change in mobility at a point in time compared with the median value of its corresponding day’s baseline recorded between 3 January and 6 February 2020, when there were no MRPs. The period considered in this study spanned from the start of the lockdown on 30 March 2020 to its easing on 5 May 2020. Mobility to six categories of places was captured in the Google Mobility report:

1. Grocery and pharmacy: grocery markets, food warehouses, farmers’ markets, speciality food shops, drug stores, and pharmacies.

2. Retail and recreation: restaurants, cafes, shopping centres, theme parks, museums, libraries, and movie theatres.

3. Parks: local parks, national parks, public beaches, marinas, dog parks, plazas, and public gardens.

4. Transit stations: public transport hubs, such as subways, bus, and train stations.

5. Workplaces: places of work.

6. Residential: places of residence.

Intra-state mobility restrictions are likely to be reliably measured by the first three categories of places. Therefore, the relative degree of movement to these locations can be used to measure the effectiveness of MRPs [23]. However, we excluded ‘parks’ from the categories used in our analysis because they could be affected by weather conditions during the period under consideration, the rainy season in Nigeria. We used the median values recorded for the Grocery and Pharmacy and Retail and Recreation categories aggregated at the state level. We then summed up the two values for each state. Any state with an aggregate value below the median cumulative value for all states was categorised as having ‘reduced mobility’, and 18 states fall in this category (Fig. 1(b)). Households in these states were assigned 1 (treated group), while the rest of the households, tagged ‘increased mobility’, were assigned 0 (control group) in our variable.

Empirical strategy for mandatory mobility restrictions

We constructed a balanced panel with the pre-pandemic face-to-face GHSP-W4 (January-February 2019) and the first round of the NLPS (April-May 2020), retaining only households with complete information in both rounds. The resulting panel had 2,015 households in both rounds. The use of pre-outbreak GHSP-W4 data, collected approximately a year before the pandemic, effectively eliminates any confounding from pre-lockdown voluntary behavioural changes or anticipatory effects, as COVID-19 had not yet emerged. We implemented the following TWFE DID specification:

1

where is a binary variable for self-reported unmet needs for households, at a round, , having values of 1 and 0 for households with and without unmet needs, respectively. and capture household fixed effects, including all time-invariant household characteristics, and time fixed effects, respectively; is a vector of household covariates (controlling for household time-variant effects), and captures the idiosyncratic error term that varies across both households and time (clustered at the enumeration area (EA) level). is a time dummy variable, assuming 0 for the pre-pandemic GHSP-W4 round period and 1 for the post-outbreak NLPS round period. is our treatment variable, which takes a value of 0 for households in the state with partial lockdown (or increased mobility) (control group) and 1 for the state with total lockdown (or reduced mobility) (treated groups) for the total-partial lockdown (or Google mobility report) approach. In this fixed-effects specification, the main effect of treatment is absorbed by the household fixed effects , as treatment status is time-invariant within households. The dummy captures the overall time trend affecting all households, and in a two-time period, =.

The coefficient, of the dummy captures the aggregate trend in our outcome variable, unmet needs. It also accounts for potential variation in our outcome driven by differences in survey methods (face-to-face or phone surveys) and survey questions. The parameter of interest is the value of the coefficient, of the interaction term between and . It captures the differential impact of mobility restrictions on households in those states with total lockdown (or reduced mobility) compared with those with partial lockdowns (or increased mobility). This is the average treatment effect on the treated households (ATET). We hypothesised that households in states with total lockdown (or reduced mobility) would experience more forgone care than their counterparts in other states. We estimated this equation using a linear probability model (LPM) and a fixed effects (FE) panel estimator, which absorbed unobserved household heterogeneity.

Impacts of voluntary self-imposed mobility restriction resulting from perceived COVID-19 risks on households’ unmet needs during the first wave of the pandemic

The probability of COVID-19 infection correlates better with the number of COVID-19 cases than with hospitalisations and deaths. Therefore, we used the cumulative number of reported cases in Nigerian states as a proxy for household perceived COVID-19 risk (Supplementary Table 3). Following Zhang, et al. (2020) [24], we classify states into four groups (Supplemental Table 3) depending on their cumulative COVID-19 confirmed cases recorded as of August 2020, using data obtained from the daily COVID-19 report by the Nigeria Centers for Disease Control (NCDC) [25]. When the cumulative number of cases is less than 100, the state is classified as low-risk; when it is between 100 and 1,000, the state is classified as medium-risk; and when it exceeds 1,000 but is less than 10,000, the state is classified as high-risk. Lagos State, the ‘epicentre’ of the COVID-19 outbreak in Nigeria, with an excess of 10,000 cases, is the fourth category (See Supplementary Fig. 4).

Empirical strategy for perceived COVID-19 risks

To estimate the differential impact of COVID-19 risks on unmet needs, we constructed a balanced panel, as we did previously and employed a DID specification:

2

where is a binary variable for unmet needs, represents a dummy for three periods with a value of 0 for the period before the lockdown (January–February 2019) and 1 for each period during the lockdown (April–May 2020) and after the lockdown (June–August 2020), and is our treatment variable, representing the state-level cumulative COVID-19 risk category discussed above. The model was estimated using LPM and a panel fixed-effect estimator; thus, and are absorbed by the household fixed effects .

Again, the coefficient, of the interaction term between each of the dummies and the time dummies, captures the aggregate differential impact of COVID-19 cases between the low-risk states and states in each of the other risk categories, as well as for each of the three time periods (‘before lockdown’, ‘during lockdown’, and ‘after lockdown’). We hypothesised that the higher the COVID-19 risk category, the higher the unmet needs experienced by households in the category compared to their counterparts in low-risk states. As before, represents household time-invariant characteristics, represents time fixed effects, represents household time-variant covariates, and is the error term clustered at the EA. For our two-time period specification is equal to .

Since the impacts of mobility restrictions and COVID-19 risks are expected to have heterogeneous treatment effects across different subgroups of households, we also examined this effect on household socioeconomic status and vulnerabilities using the expanded model:

3

Where is a binary pre-pandemic variable indicating households below the national poverty line of ₦137,430 (US$447.78) [26], households with PWNCD, or PWD. The coefficient, of the three-way interaction term(s) estimates the difference in the treatment effect across these household categories. All other parameters are as discussed before. We estimated three separate models, one for each vulnerability status, rather than including all interactions simultaneously, to avoid over-parameterisation and maintain statistical power.

Assumptions of the DID model

The DID design relies on two fundamental assumptions: exogenous shock (i.e., intervention is unrelated to the outcome at baseline) and a parallel trend assumption [27, 28]. In our case, the COVID-19 situation and lockdown measures were not anticipated, thereby fulfilling the exogenous shock assumption, which precludes any causal effect prior to the event. The parallel trend assumption presupposes that our treatment (treated and control) groups, without the lockdown/mobility restriction, would have followed the same trend concerning unmet needs (vis-à-vis utilisation). As this counterfactual situation is unobservable, it is customary to examine the situation of both groups before the shock [28]. To validate this assumption, we first demonstrated that the distributions of the time-invariant covariates were highly similar in the treated and control groups (Supplementary Tables 4 and 5). Then, we graphically demonstrated that the mean unmet needs for the treated and control groups (for mobility restriction and risks of COVID-19) moved in tandem over time before the pandemic (Supplementary Figs. 2(a, b) and 3(a–c)). Although it would have been impractical to demonstrate this with only one round of pre-treatment data, we leveraged the different interview dates in our dataset and the mean unmet needs to plot the trend graphs. To make an accurate statistical inference and correct for heteroskedasticity in our model, we used robust standard error clustered at the enumeration area as suggested in the literature [28, 29].

We used a two-way fixed effect (TWFE) for our DID analysis because there was no significant variation in treatment timing since the implementation of MRPs occurred throughout the country between 30 March and 5 May 2020 (which coincided with the time of collection of the first round of data used in our analysis), thus eliminating the bias associated with staggered or temporal variation in the implementation of MRPs when using TWFE [30, 31]. We similarly avoided this problem in the COVID-19 risk analysis, which spanned more than two time periods, by adopting spatial instead of temporal variation in classifying state-level COVID-19 risks.

All statistical analyses were performed using Stata/MP 17 for Windows (StataCorp, Texas, USA).

Results

Trend in unmet needs before, during and after the lockdown

The prevalence of needed care and unmet needs across the three time periods of the pandemic is shown in Supplementary Table 6. It shows that during the pre-pandemic period (January-February 2019), the prevalence of reported unmet needs was 16.18% (95% CI: 13.41–19.40) among households. This prevalence increased to 24.51% (95%CI: 19.74–30.00) during the lockdown (Apr-May 2020), but dropped below the pre-pandemic level to 12.51% (95% CI: 10.43–14.95) after the lockdown (Jun-Aug 2020). These trends suggest a substantial but temporary disruption to healthcare access during the lockdown period.

Impact of mobility restrictions on unmet medical needs

Figure 1(a) shows that total lockdown states were concentrated in the South-West (Lagos, Ogun, Osun, and Ekiti), North-Central (Niger, Kwara, Plateau, including the FCT), and North-East (Adamawa, Bauchi, Borno, and Taraba) zones. Figure 1(b) reveals that several northern states, which implemented total lockdowns (Adamawa, Bauchi, Borno, Plateau, Taraba), actually experienced above-median mobility, suggesting low policy compliance. Conversely, some states on partial lockdown near Lagos, such as Ogun and the FCT (including Edo, Enugu, Nasarawa, Ondo, and Oyo), experienced below-median mobility, possibly reflecting spill-over effects or voluntary restrictions due to proximity to the epicentre.

Table 1 presents DID estimates from both the partial-total lockdown approach (Approach 1) and Google Mobility approach (Approach 2). The partial-total lockdown approach shows that households in states that implemented total lockdowns experienced a 13% point (95%CI: 3–23), p < 0.05, increase in the probability of reporting unmet medical needs compared to households in partial lockdown states during April-May 2020. This represents a relative increase of approximately 102% compared to the baseline unmet needs rate of 12.72% in partial lockdown states during the same period (Supplementary Table 6). The Google mobility approach, in contrast, shows a 15% point (95%CI: 6–25), p < 0.01, increase in the probability of reporting unmet medical needs for households located in states where mobility declined below the national median value (reduced mobility) compared with those with above-median value (increased mobility). The coefficients of the dummy in both approaches are near zero and statistically insignificant, indicating no aggregate increase in unmet needs across all households (except those in states with more stringent restrictions) after controlling for the treatment group.

Table 1.

Mobility restrictions and household unmet needs in Nigeria during the April/May 2020 lockdown

Variables	Partial-Total Lockdown Methodα	Google Mobility Data Methodβ
	Unmet Needs Coefficient (95%CI)	Unmet Needs Coefficient (95%CI)
	-0.01 (-0.10–0.08)	0.00 (-0.07–0.07)
	0.13** (0.03–0.23)	0.15*** (0.06–0.25)
Household fixed effects	Yes	Yes
Observations	2,100	2,100
Number of households	1,596	1,596
R-squared	0.07	0.07

Notes: Estimations are based on Nigeria GHSP 2019 and 2020 NLPS Round 1

Post is post-pandemic dummy (2020 NLPS Round 1)

^α In the total-partial lockdown method, Treated represents households in states with a total lockdown as announced by the government

^β Google Mobility data method: Treated represents households in states with lower than median mobility (reduced mobility) in the ‘grocery and pharmacy’ and ‘retail and recreation’ categories of Google Mobility Data compared with their baseline values in Nigeria between March 30 and May 5, 2020

P values: *** p < 0.01, ** p < 0.05, * p < 0.10

95% CIs are shown in parentheses

Robust standard errors were clustered at the enumeration area, EA

Regarding mobility restrictions, there was no statistically significant difference (p < 0.05) in treatment effects on unmet needs between households below the poverty line and those above the poverty line. Similarly, no significant difference existed in the impact of mobility restrictions on unmet needs between households with and without PWNCD, or between households with and without PWD (Supplementary Tables 7 and 8).

Impact of perceived COVID-19 risks on unmet medical needs

Table 2 shows the DID estimates comparing unmet needs across COVID-19 risk categories (medium-risk, high-risk, and epicentre) with those in low-risk states, separately for the lockdown and post-lockdown periods. During the lockdown period, there was no statistically significant percentage point increase in unmet needs in states with medium risk, high risk and epicentre, compared to those with a low risk of COVID-19. After the lockdown, statistically significant effects of COVID-19 cases on unmet needs were only observed in households in Lagos State (the pandemic epicentre with > 10,000 cases), where there was a 24% point (95%CI: 2.00–46.00), p < 0.05, increase compared to households in low-risk states (< 100 cases).

Table 2.

Impacts of perceived COVID-19 risks on households’ unmet needs during the first wave of the pandemic (April-August 2020) in Nigeria

Variables	Unmet Needsβ Coefficient (95%CI)
During lockdownθ	-0.07 (-0.21–0.08)
After lockdownθ	-0.08 (-0.18–0.02)
(Medium-risk vs. Low-risk)α × During lockdownθ	0.15* (-0.01–0.31)
(High-risk vs. Low-risk)α × During lockdownθ	0.10 (-0.07–0.27)
(Epicentre vs. Low-risk)α × During lockdownθ	0.25* (-0.04–0.53)
(Medium-risk vs. Low-risk)α × After lockdownθ	0.01 (-0.10–0.12)
(High-risk vs. Low-risk)α × After lockdownθ	0.03 (-0.09–0.15)
(Epicentre vs. Low-risk)α × After lockdownθ	0.24** (0.02–0.46)
Household Fixed Effects	Yes
Number of observations	3,689
Number of households	1,509
R-squared	0.03

Notes: Estimations are based on Nigeria GHSP 2019 and 2020 NLPS Round 1

^αRepresents the treatment variable assigned with the value of 0 for households in the states with low-risk of COVID-19 infection (number of cases < 100) (as the control group) and 1 for households in the medium-risk (100 ≥ number of cases < 1,000), high-risk (1,000 ≥ number of cases < 10,000), or epicentre state (Lagos State with number of cases > 10,000) (as the treated groups)

^θRepresents three periods: before lockdown (January–February 2019), during lockdown (April-May 2020), and after lockdown (June–August 2020). In the difference-in-differences estimation, the period before lockdown is the base period

P values: *** p < 0.01, ** p < 0.05, * p < 0.10

^β95% CIs are shown in parentheses

Robust standard errors were clustered at the enumeration area (EA)

Heterogeneous effects by household vulnerability groups

Table 3 presents triple-difference estimates examining whether the impact of COVID-19 risk on unmet needs differed for vulnerable household subgroups (socioeconomic status, the presence of PWNCD and PWD). It shows that only households with PWD located in epicentre states with high COVID-19 risks demonstrated a significant increase in effect (p < 0.05). There was an increase of 40% (95%CI: 6–74) and 74% (95%CI: 22–26) points, respectively, in unmet needs in these households compared with those in low-risk states during the lockdown period. However, no significant effect was observed on unmet needs for households with PWD after the lockdown. Similarly, there was no significant differential effect of COVID-19 risks on unmet needs in households below the poverty line and those with PWNCD status either during or after the pandemic (Table 3).

Table 3.

Heterogeneity in the impact of COVID-19 risk on unmet needs between households below and above the poverty line, households with and without persons with non-communicable diseases and disability

Variables	Household Below poverty line	Household with NCDs	Household with Disability
	Unmet Needsβ Coefficient (95%CI)	Unmet Needsβ Coefficient (95%CI)	Unmet Needsβ Coefficient (95%CI)
During Lockdownθ	-0.00 (-0.15–0.15)	-0.01 (-0.19–0.17)	-0.02 (-0.19–0.15)
Status§ × (Medium-risk vs. Low-risk)α × During Lockdownθ-0.05 (-0.33–0.22)	0.31* (-0.00–0.63)	0.22* (-0.04–0.48)
Status§ × (High-risk vs. Low-risk)α × During Lockdownθ	0.01 (-0.28–0.29)	0.16 (-0.22–0.54)	0.40** (0.06–0.74)
Status§ × (Epicentre vs. Low-risk)α × During Lockdownθ	-	-0.32 (-0.81–0.16)	0.74*** (0.22–1.26)
After Lockdownθ	-0.02 (-0.15–0.11)	-0.05 (-0.17–0.07)	-0.08 (-0.20–0.03)
Status§ × (Medium-risk vs. Low-risk)α × After Lockdownθ	0.07 (-0.14–0.29)	0.11 (-0.18–0.41)	0.02 (-0.23–0.26)
Status§ × (High-risk vs. Low-risk)α × After Lockdownθ	0.08 (-0.16–0.31)	0.13 (-0.19–0.45)	-0.13 (-0.43–0.17)
Status§ × (Epicentre vs. Low-risk)α × After Lockdownθ	-	-0.23 (-0.61–0.16)	0.43 (-0.28–1.13)
Household Fixed Effects	Yes	Yes	Yes
Number of oobservations	3,689	3,689	3,689
Number of households	1,509	1,509	1,509
R-squared	0.04	0.04	0.04

Notes: Estimations are based on Nigeria GHSP 2019 and 2020 NLPS Round 1

^αRepresents treatment variable assigned with the value of 0 for households in the states with low-risk of COVID-19 infection (number of cases < 100) (as the control group) and 1 for households in the medium-risk (100 ≥ number of cases < 1,000), high-risk (1,000 ≥ number of cases < 10,000), or epicentre states (Lagos State with number of cases > 10,000) (as the treated groups)

^θRepresents three periods: before lockdown (January–February 2019), during lockdown (April-May 2020) and after lockdown (June–August 2020). In the difference-in-differences estimation, the period before lockdown is the base period

^§Represents each category of households: households below the poverty line, households with non-communicable diseases, and households with disability

P values: *** p < 0.01, ** p < 0.05, * p < 0.10

^β95% CIs are shown in parentheses

Robust standard errors were clustered at the enumeration area (EA)

Discussion

This study provides some causal evidence that government-imposed MRPs in Nigeria during the first wave of the COVID-19 lockdown substantially increased household unmet medical needs. In contrast, perceived COVID-19 risk primarily affected unmet needs in the pandemic epicentre (Lagos State) after restrictions were lifted. Using DID analyses with two complementary approaches, we found that total lockdown measures increased unmet needs by 13% points compared to partial lockdown measures, and states with below-median mobility experienced a 15% point increase in unmet needs. After the lockdown ended, households in Lagos State (the epicentre with more than 10,000 cases) experienced a 24 percentage point increase in unmet needs compared to low-risk states. Notably, households with persons with disabilities (PWD) faced disproportionate impacts during lockdown, with 40–74 percentage point increases in the epicentre.

Our findings indicate a positive causal relationship between unmet care needs and MRPs during the lockdown, showing a 13- and 15-percentage-point increase in household unmet needs between states with total and partial lockdowns, respectively, and between states with reduced and increased mobility. Perceived COVID-19 risks did not significantly impact unmet needs during the lockdown. However, after the lockdown, there was a 24%-point increase in the differential impact of COVID-19 risks on unmet needs between the epicentre of the infection (Lagos State) and states with low COVID-19 risks. The differential impacts of MRPs and COVID-19 risks on unmet needs for households with poverty, PWNCD, and PWD appear insignificant. There was, however, a negative impact of COVID-19 risks on unmet needs for households with PWD during the lockdown.

The 13–15 percentage point increases in households with unmet needs during lockdown represent more than double of baseline rates. Our study used an average estimate from mobility data, which might have underestimated the real impacts. These findings align with global evidence of pandemic-related healthcare disruptions, particularly during the lockdown period [32, 33], occasioned by government-imposed MRPs and perceived fear of COVID-19 [7, 8, 32, 34]. However, the convergence of estimates from our two approaches (lockdown stringency: 13%; mobility reduction: 15%) provides good evidence that government-mandated policies, rather than voluntary behavioural SMR, drove the majority of unmet needs in Nigeria during lockdown, consistent with studies in Nigeria and other LMICs during the early pandemic [19, 32, 35, 36].

These findings contradict reports from the US and Europe, where voluntary risk avoidance often preceded and exceeded the impacts of policy changes, without a significant association between the MRPs and unmet needs [11, 37]. High-income countries experienced sustained reductions in healthcare usage for months beyond lockdown, while Nigeria saw a rapid normalisation, with unmet needs declining to 12.5% after lockdown. These disparities may result from the severe indirect impact of mandated restrictions on Nigeria’s fragile healthcare system, compared to the more resilient healthcare systems of high-income countries. Mobility restrictions in high-income countries, which faced higher COVID-19 case numbers and mortality, were likely driven by the perceived risk of infection rather than mandated restrictions [38]. The temporary nature of the impacts in Nigeria suggests that access disruption, rather than resource depletion, was the primary mechanism at play. If the healthcare system had suffered lasting damage, such as health worker attrition, facility closures, or a supply chain collapse, the effects would likely have persisted or worsened after the lockdown. Although there is evidence that mandated restrictions may have helped contain COVID-19 infections in Nigeria [35], the negative impacts of MRPs on unmet needs cannot be overlooked.

The discordance between announced policies and actual mobility reveals important realities of implementation. Northern states with total lockdown showed high mobility, suggesting limited enforcement capacity or economic necessity overriding compliance [36, 39]. Moreover, Southern states near Lagos with partial lockdown showed reduced mobility, indicating spill-over effects from proximity to the epicentre or greater voluntary compliance. Studies have shown that the spill-over effect of MRPs on neighbouring communities could be substantial [40, 41].

The shifting role of COVID-19 risk across time periods illuminates the interplay between policy, compliance and perception. During lockdown, risk perception had minimal additional impact beyond MRPs, as the constraint was access rather than avoidance. However, after lockdown, risk perception became dominant only in the epicentre, where media saturation, an increased number of cases, and healthcare system overwhelm created sustained deterrence, and extreme case burdens (> 10,000) translated to sustained behavioural change after policy barrier lifting [33, 42, 43].

The impacts of lockdown measures and COVID-19 risks on unmet needs for people with socioeconomic and health vulnerability, such as the poor and those with PWNCD and PWD, have been established [11, 33]. Unexpectedly, except for PWD during the lockdown, our findings showed no significant differential impact between those households and their counterparts. These findings align with the predominant impact of government-imposed MRP, which affected all households regardless of socioeconomic, health or disability status. However, poor households might have been indirectly affected by the ensuing socioeconomic hardship [44]. Our findings may indicate that households with PWNCD prioritised the risk of severe outcomes from NCD complications over the risk of COVID-19 infection (which was minimal anyway), as observed in Korea, where there was no decrease in the use of healthcare services by people with severe diseases [45].

Nevertheless, it is vital to recognise that any unmet needs for PWNCD portend grave aftermaths [46]. For example, approximately 84% of PWNCD in Nigeria experienced deterioration in their health conditions due to difficulty accessing their medicines during the pandemic [34]. Moreover, aligning with other studies, our study shows PWD were differentially affected by unmet needs during the lockdown period [47]. This effect was likely due to the impact of MRPs on transportation, particularly specialised transport like wheelchair-accessible vehicles and assistance from caregivers who may have been unavailable due to their own movement restrictions. Since NCDs are major drivers of chronic disabilities [48], PWNCD would have experienced a chunk of the impact on PWD.

Practical implications and contributions

Our findings make a significant contribution to the literature and carry several actionable implications for pandemic preparedness in Nigeria and similar resource-limited settings. Our study expands the literature on the indirect impact of the COVID-19 pandemic on unmet needs and healthcare utilisation. Beyond substantive findings, this study advances methodology for pandemic impact assessment. We demonstrate a practical approach to disentangling the effects of government policy from voluntary behavioural responses by utilising spatial variation in policy stringency and objective mobility data. It contributes to the raging debate on the harm versus the benefits of MRPs during the pandemic. With approximately three times more deaths from non-COVID causes than from COVID-related deaths between 2020 and 2021 [1], the health opportunity cost of MRPs might be unjustifiably enormous in countries like Nigeria with low infection or fatality rates [49], where nonrestrictive containment measures, such as handwashing and face masking, may have been more advantageous. Therefore, future pandemic responses should focus on implementing restrictions in high-transmission areas while allowing mobility in low-risk regions. This approach should utilise real-time monitoring of COVID-19 trends and healthcare utilisation to guide policy changes.

Additionally, health policy should prioritise the care of vulnerable population through maintaining supply chains for essential medicines, employing appointment scheduling, health worker home visits, and alternative care delivery methods, such as mobile clinics, community pickup points, and pharmacy-based services for high-risk patients, and deploying telemedicine for stable chronic disease patients [50, 51]. The impacts on households with PWD necessitate proactive inclusion in pandemic preparedness. This includes maintaining specialised disability services such as rehabilitation, mental health support, and the provision of assistive devices; ensuring accessible communication systems; providing accessible transport for PWD to reach healthcare facilities; and developing emergency response protocols for disability-related needs [52, 53].

Limitations

The causal relationship in our study should be interpreted with caution, considering some potential methodological and data limitations. First, unmet needs were self-reported and retrospective, covering a period of 1 to 3 months, which are subject to recall and social desirability biases. However, the self-reported unmet needs measure has been widely validated in health services research and extensively used to assess healthcare access during the COVID-19 pandemic [33]. Moreover, the use of a dichotomous unmet need indicator (yes/no) does not capture the severity, frequency, and type of care missed. Future research should utilise continuous measures (e.g., number of forgone visits, days delayed) or severity-weighted indices.

Second, the NLPS data were collected via phone interviews, drawn from the pre-existing GHSP panel, which had used face-to-face interviews. The phone interview could introduce selection bias by excluding households without phone access or creating mode effects, where phone surveys may encourage shorter or inaccurate responses due to privacy concerns. Potential selection bias was mitigated by the survey strategy, which ensured the collection of phone numbers from most households interviewed in GHSP-W4, and by initiating multiple calls to them during NLPS interviews, resulting in response rates exceeding 90%. Moreover, our DID design accounted for any systematic mode effects.

Third, one of the challenges of using DID in COVID-19 studies is the spatiotemporal variability in the adoption of MRPs, as these policies are usually adjusted in response to the severity or risk of infection, as seen in Rwanda and the US [54]. This poses an endogeneity problem, as it is challenging to disentangle the effects of MRPs from those of COVID-19. Fortunately, unlike Rwanda and the US, Nigeria maintained a sustained MRP level during the lockdown [54]. We took advantage of the only heterogeneity (partial or total) that exists in the MRPs for our analysis, eliminating the so-called ‘treatment timing’ effect [17]. This suggests that our findings may not be generalisable to some countries and outside the 36-day lockdown period in Nigeria.

Fourth, despite the DID design controlling for time-invariant confounders, time-varying confounders could still bias the estimates; however, our TWFE model controlled for such biases. The similarity in estimates from lockdown stringency and mobility approaches also suggests minimal confounding from omitted time-varying factors. Although the use of state-level total versus partial lockdowns may mask within-state variation in treatment assignment, potentially attenuating estimates if compliance varied widely, the convergence with the mobility approach suggests that this effect may be minimal.

Fifth, our proxy for perceived COVID-19 risks relied on reported cumulative cases instead of actual infection rates (usually estimated using the reproduction number, R [55]), which may underestimate the true infection burden due to Nigeria’s limited testing capacity during the early phase of the pandemic. This could bias our estimates of risk-related impacts on unmet needs, potentially understating the role of voluntary SMRs in low-testing areas. However, it is most probable that people would respond to reported cases rather than the non-apparent infection rate.

Conclusion

Despite these limitations, our findings provide robust insights that MRPs, while potentially reducing COVID-19 transmission, created substantial healthcare access barriers affecting millions of households in Nigeria. The disproportionate impact on persons with disabilities highlights systemic gaps in disability-inclusive health emergency preparedness. This accentuates the need for a tailored control policy, based on risk assessment levels, that protects both against novel infectious threats and the disruption of essential healthcare for endemic conditions, as well as targeted interventions for vulnerable populations, to minimise adverse health consequences during future pandemics. The evidence offers both cautionary lessons about the unintended consequences of well-intentioned policies and optimistic insights into health system resilience and the potential for better-designed interventions in future emergencies.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ATET

Average treatment effect on the treated

CI

Confidence interval

COVID-19

Coronavirus disease 2019

DID

Difference-in-differences

EAs

Enumeration areas

FCT

Federal Capital Territory

GHSP

General Household Survey Panel

LCC

Limited Liability Company

LGA

Local government areas

MRPS

Mobility-restricting policies

NBS

Nigeria National Bureau of Statistics

NCDC

Nigeria Centers for Disease Control

NCDs

Non-communicable diseases

NLPS

National longitudinal phone surveys

PWD

Persons with a disability

PWNCD

Persons with non-communicable diseases

RCTs

Randomised control trials

RE

Random effect

SARS-CoV-2

Severe acute respiratory syndrome-coronavirus-2

SMRs

Self-imposed mobility restrictions

SSA

Sub-Saharan Africa

TWFE

Two-way fixed effect

US

United States

WB

World Bank

WHO

World Health Organization

Author contributions

AO and KA conceptualised and designed the study. AO analysed and interpreted the data. AO drafted the manuscript and received critical feedback from HS and KA. All authors (AO, HS and KA) read and approved the final version of the manuscript.

Funding

This study is part of PhD research, and no financial support was received for this article’s authorship and publication.

Data availability

The data supporting this study’s findings are available in the World Bank microdata library at https://microdata.worldbank.org/index.php/catalog/3557/get-microdata; https://microdata.worldbank.org/index.php/catalog/3712/get-microdata.

Declarations

Ethical approval and consent to participate

Ethical approval for this research was waived by Murdoch University Human Research Ethics Committee (project approval reference number 2022/159) because only anonymous secondary data, which are publicly available, were used. More details regarding NBS data policy and ethical standards are available at https://nigerianstat.gov.ng/page/data-privacy-policy.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.