The impact of artificial intelligence (AI) on maternal mortality: evidence from global, developed and developing countries

Abstract

Background

This study examines the impact of Artificial Intelligence (AI) on maternal mortality in alignment with Sustainable Development Goal (SDG) 3.1, which aims to reduce maternal mortality to below 70 per 100,000 live births by 2030. Despite advancements, maternal mortality remains disproportionately high in developing countries due to weaker healthcare infrastructure.

Methods

Using panel data from 70 countries (1990–2022), sourced from WHO’s Global Burden of Disease (GBD), World Bank’s World Development Indicators (WDI), UNCTAD, and the World Robotics database, we apply the Difference-in-Differences (DiD) approach to assess AI’s impact over time and the Auto-Regressive Distributed Lag (ARDL) model to examine short- and long-term effects.

Results

AI adoption significantly reduces maternal mortality, particularly in developing countries, where post-2000 advancements have led to notable declines. ARDL results show that 27% of deviations from long-term maternal mortality trends are corrected annually, highlighting AI’s sustained impact. The DiD analysis indicates AI’s greatest benefits in resource-limited settings, including improving early diagnostics, personalized care, and remote monitoring. In developed countries, AI’s effects are marginal due to existing advanced healthcare systems.

Conclusion

AI presents a transformative solution for reducing maternal mortality, particularly in low-resource settings. Policymakers should prioritize AI-driven healthcare, expand digital infrastructure, and ensure equitable access to maximize its benefits. AI integration is crucial for addressing maternal health disparities and accelerating progress toward SDG 3.1.

Introduction

Maternal mortality remains a significant global health challenge, with substantial disparities between developed and developing countries. Despite a 34 percent decline in maternal mortality between 2000 and 2020, progress has been uneven, with low- and middle-income countries (LMICs) continuing to experience disproportionately high mortality rates [1]. Sustainable Development Goal (SDG) 3.1 aims to reduce maternal mortality to fewer than 70 deaths per 100,000 live births by 2030. However, many countries remain far from achieving this target due to persistent structural barriers, including inadequate healthcare infrastructure, limited access to skilled medical personnel, financial constraints, and disparities in maternal healthcare quality [1]. Maternal mortality is often linked to preventable complications such as postpartum hemorrhage, eclampsia, obstructed labor, and sepsis—conditions that could be mitigated through early detection and timely intervention. However, healthcare systems in resource-limited settings often lack diagnostic tools, trained personnel, and efficient monitoring systems, exacerbating maternal health risks [2]. Addressing these challenges requires innovative, scalable, and cost-effective solutions to strengthen maternal healthcare delivery, particularly in high-burden regions.

Artificial Intelligence (AI) is increasingly recognized as a transformative tool in healthcare, offering data-driven insights and automation to enhance maternal health outcomes. AI-powered predictive analytics can identify high-risk pregnancies, enabling earlier interventions that reduce preventable maternal deaths [3]. Machine learning algorithms improve diagnostic accuracy, detect complications, and assist healthcare professionals in making evidence-based decisions, mainly where traditional methods fall short [4]. AI-driven decision-support systems analyze large datasets to predict maternal complications, allowing healthcare providers to allocate resources more effectively [5]. In developing countries, where limited access to skilled maternal healthcare providers remains a critical challenge, AI-powered telemedicine, and remote monitoring solutions can help bridge the gap by offering virtual consultations, continuous maternal health surveillance, and automated risk assessment [6]. Natural language processing (NLP) and AI chatbots have also provided maternal health education and personalized pregnancy guidance, improving maternal health literacy and self-care practices [7].

Despite its significance, the empirical evidence on AI’s effectiveness in reducing maternal mortality remains limited, particularly in developing countries, where healthcare systems are less equipped to integrate AI-based interventions [8]. While AI has been widely studied in high-income countries with advanced healthcare infrastructure, little research has explored its impact in LMICs, where healthcare system deficiencies, digital divides, and low AI adoption rates may limit its effectiveness. AI implementation is further constrained by inadequate digital infrastructure, limited technical expertise, data privacy concerns, and potential algorithmic biases that may reinforce existing healthcare inequalities [9]. Moreover, while AI technologies have demonstrated promising results in pilot studies and controlled environments, their real-world impact on reducing maternal mortality at scale remains underexplored. Given these disparities, empirical research is urgently needed to evaluate AI’s role in maternal healthcare across diverse socioeconomic contexts, focusing on how AI can be tailored for resource-limited settings.

This study addresses this gap by investigating the impact of AI adoption on maternal mortality across 70 countries, specifically focusing on differences between developed and developing nations. The study makes three main contributions. First, it contributes to the empirical literature on the field by integrating AI variables (AI stock and AI flow) with maternal health outcomes to offer unique insights into AI’s impact over time and across different economic contexts. Secondly, using AI robotics data from the World Robotics database (1990–2022) and employing econometric techniques such as fixed effects regression (and its forecasting components), Difference-in-Differences (DiD, and Auto-Regressive Distributed Lag (ARDL) models, we assess both the short- and long-term effects of AI on maternal mortality. These methods allow for a nuanced understanding of the temporal and spatial variations in the data. These diverse methodologies provide a robust framework for overcoming limitations seen in earlier studies, such as addressing endogeneity issues and capturing complex dynamics that simpler models may overlook. Specifically, the dynamic panel ARDL approach provides flexibility by allowing the examination of both long-term relationships and short-term dynamics without the strict requirement of stationary series, making it ideal for diverse economic and healthcare environments [10, 11]. Meanwhile, the DiD approach strengthens the analysis by isolating the effects of AI implementation from other confounding factors, enhancing the robustness of the analysis [12]. These methodologies collectively enable a sophisticated analysis that adds depth to understanding how AI technologies can be strategically leveraged to enhance maternal health outcomes globally.

Thirdly, the study links Grossman’s health capital model with modern AI applications [13], enriching the theoretical discourse on technology’s role in enhancing health capital. It posits that AI serves as an investment in health capital, potentially reducing maternal mortality through improved healthcare services and patient outcomes. Lastly, this study is critical for projecting AI’s transformative potential in global healthcare delivery and maternal mortality outcomes. By providing empirical evidence, it supports integrating AI more widely into healthcare policies and practices, particularly in developing countries. By elucidating these contributions, our research fills a critical gap in the existing literature. It sets the stage for future studies to explore other dimensions of AI’s impact on different aspects of public health.

The remainder of this paper is structured as follows: Section 2 outlines the methodology and data sources, Section 3 presents the empirical findings, and Section 4 discusses conclusion and policy implications.

Methodology and data

Empirical strategies

To achieve the objectives of this study, we employed descriptive statistics to clearly and concisely summarize the main features of the dataset. Additionally, we applied the pairwise correlation approach to explore initial relationships between the variables. This section outlines other analytical methods, including the spatial visual approach, difference-in-difference estimator, fixed effects regression (and its forecasting components), and the dynamic panel autoregressive distributed lag (ARDL) approach. Detailed explanations of these methods are provided in the following subsections.

Theoretical underpinning and empirical model specifications

This study draws on the demand for health theory, which suggests that health investments, including technological innovations, affect overall health outcomes. The concept of"health capital,"introduced by Grossman, highlights that health can improve or deteriorate over time through investments such as medical care and preventive measures [13]. Integrating AI into maternal health strategies can enhance the effectiveness of these investments, reduce barriers, and lower maternal mortality rates. The study also considers socioeconomic determinants influencing mortality, aligning them with the demand for health framework. Furthermore, previous studies have shown that AI can enhance maternal healthcare through advanced diagnostics, personalized treatment plans, and predictive analytics [4, 14]. AI algorithms analyze medical data to predict high-risk pregnancies and suggest early interventions, potentially reducing maternal mortality rates [15]. For example, AI-driven models can forecast complications like preeclampsia and gestational diabetes more accurately than traditional methods, facilitating timely and targeted care [16]. Therefore, the functional form equation can be presented as 1:

  

$$mm{r}_{\mathit{it}}=f(A{I}_{\mathit{it}},X_{\mathit{it}})$$ 1

where $\mathit{mmr}$ is maternal mortality, $\mathit{AI}$ represents overall AI application in healthcare, and $X$ includes other explanatory variables influencing maternal mortality, such as healthcare access, infrastructure quality, socioeconomic factors, and health policy environments. $\mathit{iandt}$ represent individual country and time, respectively. Further details on the control variables are provided in Table 1. Before expanding on Eq. 1, it is important to note that the overall AI index, generated through Principal Component Analysis, comprises two key indicators: AI Stock and AI Flow. AI stock represents the accumulated AI technologies within healthcare systems, enhancing monitoring and intervention in maternal health, thus reducing mortality [3]. AI flow involves continuously integrating new AI innovations, enabling healthcare providers to respond more effectively to emergencies and improving survival rates [17]. The functional form of Eq. 1 is further subdivided into two parts, as detailed in the footnote.

Table 1.

Summary statistics

Variable	Obs	Mean	Std. Dev.	Min	Max	N	Mean	SD	Min	Max	N	Mean	SD	Min	Max
	All countries					Developed					Developing
mmr	1988	55.957	157.266	1.082	1792.347	1092	12.033	12.062	1.082	89.765	896	109.49	222.507	4.519	1792.347
aiapflow	1988	2094.331	9534.133	0	175,546	1092	2673.549	7448.472	0	55,240	896	1388.41	11,544.252	0	175,546
aiapstock	1988	16,068.02	62,002.218	0	956,477	1092	23,941.954	67,882.369	0	412,961	896	6471.663	52,437.923	0	956,477
cm ind	1360	-.855	.836	−2.618	2.156	760	−1.243	0.514	−2.618	.346	600	-.364	0.906	−2.064	2.156
cm apw	1360	19.687	9.726	7.3	54.2	760	14.959	5.527	7.3	29	600	25.676	10.566	7.9	54.2
cm srf	1400	5.77	3.826	.7	24.4	780	7.305	3.677	.7	24.4	620	3.84	3.060	.7	16.4
cm hivpf	1953	27.58	11.954	6.1	85.59	1085	25.359	9.314	7.48	50.34	868	30.358	14.119	6.1	85.59
ha ind	1270	.621	1.147	−1.91	4.045	844	1.18	0.805	−1.23	4.045	426	-.487	0.893	−1.91	1.984
ha nmt	1366	6.832	4.263	.14	23.07	884	8.618	3.760	1.09	23.07	482	3.557	3.003	.14	13.34
ha ppt	1478	2.678	1.183	.02	7.06	955	3.275	0.857	1.01	7.06	523	1.59	0.885	.02	4.26
he ind1	1451	.901	1.944	−1.585	11.124	819	1.815	1.987	−1.043	11.124	632	-.283	1.040	−1.585	5.246
he gdp	1452	6.971	2.72	1.6	19.69	819	8.318	2.157	3.86	18.76	633	5.228	2.359	1.6	19.69
he pc	1452	1767.78	2089.294	13.21	11,758.42	819	2645.399	2272.194	21	11,758.42	633	632.283	1024.728	13.21	6467
he ppp	1451	2037.138	1782.02	52.35	11,758.42	819	2786.383	1861.323	149	11,758.42	632	1066.203	1064.540	52.35	6434
he oops	1452	28.8	15.433	5.21	85.05	819	23.617	10.092	7.14	51.94	633	35.507	18.308	5.21	85.05
inf ind	1901	.743	.89	−2.306	2.232	1075	.996	0.793	-.358	2.221	826	.414	0.903	−2.306	2.232
inf ael	1923	96.61	10.951	7.7	100	1092	99.656	1.419	88.1	100	831	92.607	15.711	7.7	100
inf inp	1966	39.38	32.963	0	100	1075	48.318	33.113	0	99.53	891	28.596	29.375	0	100
se indp	179	-.725	1.275	−3.191	2.473	27	-.619	1.195	−1.827	1.301	152	-.744	1.291	−3.191	2.473
se gdppc	1958	21,013.924	19,489.455	354.09	87,123.66	1090	27,099.931	19,484.684	740.63	87,123.66	868	13,371.358	16,597.229	354.09	73,493.27
se hwf	219	75.457	22.241	10.72	98.13	27	94.114	5.150	86.98	98.13	192	72.833	22.468	10.72	97.4
se eprf	1608	45.892	12.895	7.62	85	999	49.145	9.699	19.95	85	609	40.557	15.466	7.62	71.89
se flfpr	1988	56.473	16.209	10.66	86.25	1092	64.451	9.617	32.75	86.25	896	46.749	17.269	10.66	80.11
se indn	894	-.772	1.137	−2.27	7.482	623	−1.2	0.413	−1.753	2.871	271	.21	1.580	−2.27	7.482
se phr2 15	1060	3.772	9.026	0	81.5	715	.929	2.725	0	33.5	345	9.664	13.555	0	81.5
se phr3 65	1060	8.975	17	0	94.7	715	2.331	6.811	0	66	345	22.743	22.617	0	94.7
se phr6 85	1060	18.87	26.268	0	99.1	715	6.734	14.057	0	90.8	345	44.021	27.813	0	99.1
se unp	1360	5.042	5.722	2.5	50.4	780	2.945	2.871	2.5	33.4	580	7.863	7.201	2.5	50.4
se adr	1988	51.339	10.721	16.17	92.09	1092	49.539	5.226	36.48	70.94	896	53.532	14.599	16.17	92.09
de frt	1988	2.034	.851	.84	6.57	1092	1.588	0.333	.84	3.11	896	2.577	0.968	.92	6.57
oca bsw	1267	96.089	10.639	18	100	775	99.159	1.192	88.9	100	492	91.254	15.852	18	100

Source: Author’s Computation

The explicit functional form of Eq. 1 is as follows:

$${\mathit{mmr}}_{\mathit{it}}={\mathit{AI}}_{\mathit{it}}+X_{\mathit{it}}$$ 2

The econometric model is specified as follows:

$${\mathit{mmr}}_{\mathit{it}}={\beta }_0+{\beta }_1{\mathit{AI}}_{\mathit{it}}+{\sum }_{i=1}^9{\beta }_{2i}{X}_{i,t}+{\mathrm{ϵ}}_{1i,t}$$ 3

where: ${\mathrm{ϵ}}_{\mathit{it}}$ is the error term; ${\beta }_0$, ${\beta }_1$, and ${\beta }_2$ are parameters to be estimated. The framework above offers a comprehensive foundation for analyzing AI’s impact on maternal mortality by integrating technological advancements with traditional health determinants to evaluate their combined effect on maternal health outcomes. The econometric techniques discussed in the following sections are derived from Eq. 4.

Difference-in-Difference (DiD) econometric specification

The Difference-in-Differences (DiD) econometric technique is ideal for our study of the impact of AI on maternal mortality across countries for several reasons. First, health policy interventions are not implemented simultaneously worldwide, making DiD effective for analyzing treatments like AI in healthcare introduced at specific times and locations. It allows us to compare changes over time between treatment and control groups [18, 19]. Secondly, we used DiD methods with country-specific fixed effects to account for observed and unobserved heterogeneities at the country level. Additionally, DiD controls for confounding variables and is robust to external shocks affecting all groups [19].

The basic DiD model estimated is specified as follows:

$${\mathit{mmr}}_{\mathit{it}}={\beta }_0+{\beta }_1{\mathit{Post}}_t+{\beta }_1{\mathit{Treatment}}_i+{\beta }_2\left({\mathit{Treatment}}_i,\times ,{\mathit{Post}}_t\right)+{\phi }_i+{\phi }_t+{\mathrm{ϵ}}_{i,t}$$ 4

where ${\mathit{Treatment}}_i$ takes 1 if having non-zero AI robots flows or stocks and 0 otherwise; and ${\mathit{Post}}_t$ takes 1 if year $\ge$ 2000 and 0 otherwise; ${\phi }_i$ is fixed effects for countries; and ${\phi }_t$ is the time effect capturing global or common shocks affecting all countries.

Dynamic panel autoregressive distributed lag (ARDL) approach

This study employed the dynamic panel ARDL approach to assess AI’s impact on maternal mortality for several reasons: it handles both short- and long-term dynamics, accommodates mixed integration orders, and allows for causality and long-term relationships. Additionally, it offers flexibility, as it does not require all variables to be stationary, unlike methods such as the Engle-Granger two-step procedure [16, 20, 21]. Therefore, the panel ARDL is formulated as follows:

$$\mathrm{\Delta }{\mathit{logmmr}}_{\mathit{it}}={\beta }_0+{\sum }_{i=1}^k{\beth }_{\mathit{ij}}\mathrm{\Delta }{\mathit{logmmr}}_{j,t-i}+{\sum }_{i=1}^k{\beta }_{\mathit{ij}}{log\mathrm{\Delta }X}_{i,t}+{\alpha }_1{\mathit{logmmr}}_{j,t-i}+{\sum }_{i=1}^k{\alpha }_{\mathit{ij}}{\mathit{logX}}_{i,t}+{\mathrm{ϵ}}_{i,t}$$ 5

In Eq. 7, $i=1,...,n$ is the country index, $t=1,...,T$ is the time index, ${\mathrm{ϵ}}_{i,t}$ is the error term, $\mathit{mmr}$ is maternal mortality, and $X$ includes all the explanatory variables influencing maternal mortality (please see Table 1), $\mathrm{\Delta }$ is the first variation factor, and $k$ is the ideal lag length. To investigate the long-term cointegration relationship between the variables, the following assumptions are made:

$$H_0={\mathrm{\Omega }}_1={\mathrm{\Omega }}_2,\cdots ,{\mathrm{\Omega }}_n=0(\mathrm{There}\mathrm{is}\mathrm{no}\mathrm{cointegration})$$ 6

$$H_1=\mathrm{\Omega }\ne {\mathrm{\Omega }}_2,\cdots ,{\mathrm{\Omega }}_n\ne <spanclass{=}^{\prime }crossLinkCiteEq{u}^{\prime }>0</span>(\mathrm{There}\mathrm{is}\mathrm{cointegration})$$ 7

The assumption of no cointegration can be tested and compared with cointegration using the F test, which applies regardless of whether the variables are I(0), I(1), or a combination of both. Given the small sample size, the analytical approach found in previous studies was applied [20, 22]. The test uses panel autoregressive distributed lag bounds. If the F statistic exceeds the I(1) bound, cointegration exists; if below I(0), we accept the null hypothesis. If in between, no clear conclusion is drawn. Once a long-run relationship is established between the dependent variables and the regressors, the panel ECM model, as shown in Eq. (9), can be expressed as follows:

$$\mathrm{\Delta }{\mathit{logmmr}}_{\mathit{it}}={\beta }_0+{\sum }_{i=1}^k{\beth }_{\mathit{ij}}\mathrm{\Delta }{\mathit{logmmr}}_{j,t-i}+{\sum }_{i=1}^k{\beta }_{\mathit{ij}}{log\mathrm{\Delta }X}_{i,t}+{\alpha }_i{\mathit{ECM}}_{i,t-1}+{\mathrm{ϵ}}_{i,t}$$ 8

The coefficient ${\alpha }_i$ in the ECM represents the speed at which adjustments are made annually toward long-run equilibrium.

Fixed effect regression multivariate forecasting approach

Fixed effect regression in a multivariate forecasting context involves controlling for individual-specific characteristics that do not change over time and could bias the estimated coefficients if not considered. This model is especially useful in panel data settings where the same entities (such as countries, companies, or individuals) are observed over multiple periods [23]. In multivariate forecasting, the fixed effect model can be expressed as follows:

$${\mathit{mmr}}_{\mathit{it}}={\beta }_0+{\beta }_1{\mathit{AI}}_{\mathit{it}}+{\sum }_{i=1}^9{\beta }_{\mathit{ki}}{X}_{i,t}+{\mu }_i+{\mathrm{ϵ}}_{i,t}$$ 9

where:

• ${\mathit{mmr}}_{\mathit{it}}$ is the dependent variable for country $i$ at time $t$.

• $X_{i,t}$ is the vector of the explanatory variables for country $i$ at time $t$.

• ${\beta }_0$, ${\beta }_1$,…, ${\beta }_k$ are the coefficients to be estimated.

• ${\mu }_i$ represents the unobserved individual-specific effect (fixed effect) that captures all time-invariant characteristics of each entity.

• ${\mathrm{ϵ}}_{i,t}$ is the error term for country $i$ at time $t$, assumed to be independently and identically distributed.

For the forecast element

The forecast element in this model involves predicting future values of ${\mathit{mmr}}_{\mathit{it}}$ based on known or forecasted values of the $X_{i,t}$ variables, including AI. After estimating the coefficients $\beta$ and the individual fixed effects ${\mu }_i$, the model can be used for forecasting by inputting the values of $X_{i,t}$ for future periods:

$$\widehat{{\mathit{mmr}}_{\mathit{it}}}={\widehat{\beta }}_0+{\widehat{\beta }}_1{\mathit{AI}}_{\mathit{it}}+{\sum }_{i=1}^k{\widehat{\beta }}_{\mathit{ki}}{X}_{i,t}+{\mathrm{ϵ}}_{i,t}$$ 10

where $\widehat{{\mathit{mmr}}_{\mathit{it}}}$ is the forecasted value of ${\mathit{mmr}}_{\mathit{it}}$ for country $i$ at future time $t$.

Data

This study employs annual panel data from 70 countries, disaggregated into 39 developed and 31 developing nations, sourced from multiple global databases, including the World Robotics Database (1990–2022), WHO’s Global Burden of Disease (GBD), World Bank’s World Development Indicators (WDI), and UNCTAD. The selection of countries was guided by data availability and relevance to the research objectives. A rigorous data cleaning and harmonization process was implemented to ensure consistency and comparability across the 70-country dataset. Given the variations in data reporting standards, coverage, and definitions across sources such as GBD, WDI, and UNCTAD, systematic procedures were applied to align variables. First, differences in measurement units and reporting methodologies across countries were addressed through rescaling and normalization techniques to ensure uniformity. Second, Principal Component Analysis (PCA) was employed to construct composite indices for AI, healthcare access, and socioeconomic factors, reducing dimensional inconsistencies while enhancing cross-country comparability. Finally, countries were classified as “developed” and “developing” countries based on internationally recognized categorizations, ensuring robustness in subgroup analyses. These steps enabled a harmonized dataset, facilitating a meaningful comparative analysis of maternal mortality and AI adoption trends across diverse healthcare systems. The Table 9 in Appendix 1 outlines the variables used in this research. The list of the countries can be found in Table 10 in Appendix 2.

Results

Descriptive statistics

Table 1 presents the descriptive statistics across all countries and separately for developed and developing nations. The global maternal mortality ratio (MMR) averages 55.96 deaths per 100,000 live births, with substantial variation (Std. Dev. = 157.27), reflecting disparities in healthcare access and maternal health outcomes. Developed countries report a significantly lower average MMR of 12.03 deaths per 100,000 live births (Std. Dev. = 12.06), suggesting minimal variation and relatively uniform healthcare quality. In contrast, developing countries experience substantially higher maternal mortality, averaging 109.49 deaths per 100,000 live births (Std. Dev. = 222.51), with more significant variation. This wide disparity may indicate inconsistent maternal healthcare access and higher risk factors in resource-limited settings.

Globally, the adoption of AI-driven robotics in maternal healthcare remains uneven. The average flow of industrial and service robots is 2,094.33 units, while the stock (cumulative AI adoption) reaches 16,068.02 units, with significant cross-country variation (Std. Dev. = 9,534.13 and 62,002.22, respectively). Developed countries exhibit higher AI integration, with an average robot flow of 2,673.55 units and a stock of 23,941.95 units, reflecting advanced digital healthcare infrastructure and greater technological investments. However, developing countries show substantially lower AI adoption, with an average robot flow of just 1,388.41 units and a stock of 6,471.66 units, indicating a lag in AI-driven healthcare innovation and limited technological resources.

Key indicators that reduce maternal mortality, such as healthcare access and socioeconomic indicators, are significantly lower in developing countries, while deprivation factors remain high. The low levels of AI adoption, insufficient infrastructure, and weak health systems suggest that AI-driven maternal health improvements could be particularly impactful in developing countries despite potential labor market challenges [24].

The pairwise correlation results in Table 2 reveal several significant relationships between AI adoption, healthcare factors, and maternal mortality rates. The negative correlation between industrial and service robotics (flow and stock) and MMR suggests that advancements in AI technology contribute to lower maternal mortality rates. AI technology improves maternal health outcomes through enhanced diagnostics, efficient treatment methods, streamlined healthcare processes, and better resource allocation [25, 26]. However, inadequate healthcare infrastructure and limited human capital investments may hinder AI’s effectiveness in reducing maternal mortality, particularly in resource-constrained settings [27].

Table 2.

Pairwise correlations

Variables	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
(1) aiapflow	1.000
(2) aiapstock	0.913a	1.000
(3) mmr	−0.056 a	−0.069 a	1.000
(4) ha_ind	−0.031	0.020	−0.424 a	1.000
(5) he_ind1	0.155a	0.267a	−0.226 a	0.765a	1.000
(6) he_oops	−0.073 a	−0.142 a	0.446a	−0.478 a	−0.559 a	1.000
(7) cm_ind	−0.186 a	−0.254 a	0.555a	−0.453 a	−0.520 a	0.442a	1.000
(8) oca_bsw	0.079a	0.082a	−0.701 a	0.377a	0.250a	−0.474 a	−0.382 a	1.000
(9) se_indn	−0.065 a	−0.099 a	0.759a	−0.508 a	−0.436 a	0.593a	0.604a	−0.665 a	1.000
(10) se_indp	0.198a	0.216a	0.253a	−0.081	0.014	−0.227 a	−0.013	0.245a	−0.241 a	1.000
(11) de_frt	−0.130 a	−0.160 a	0.668a	−0.415 a	−0.296 a	0.412a	0.589a	−0.694 a	0.681a	−0.243 a
(12) inf_ind	0.146a	0.176a	−0.523 a	0.495a	0.626a	−0.589 a	−0.627 a	0.421a	−0.714 a	−0.123

Source: Author’s Computations

^a denotes the level of significance at 10% respectively

Healthcare access and health expenditure also play a crucial role. The health access index and health expenditure correlate negatively with MMR, confirming their importance in reducing maternal mortality [28, 29]. In contrast, out-of-pocket (OOP) health expenditure positively correlates with MMR, indicating that higher personal healthcare costs limit access, particularly for low-income populations, leading to poorer maternal health outcomes [30]. The negative correlation between OOP and overall health indicators supports the argument that greater financial burdens reduce healthcare accessibility.

Pre-existing health conditions significantly influence maternal mortality, as shown by the positive correlation between comorbidities and MMR. This finding underscores the heightened risks faced by pregnant women with underlying health conditions, aligning with previous research on maternal health vulnerabilities [2, 31]. Additionally, key socioeconomic and healthcare factors affect maternal mortality rates. Greater access to obstetric care, improved infrastructure, and affluent socioeconomic conditions negatively correlate with MMR, reinforcing their role in improving maternal health outcomes [32, 33]. Conversely, poor socioeconomic conditions and higher fertility rates are positively correlated with MMR, indicating that economic hardship and high birth rates contribute to increased maternal mortality.

The key determinants of maternal mortality are strongly correlated with each other and, more importantly, with AI robotics indicators, raising concerns about multicollinearity in regression models. To address this, the subsequent analysis emphasizes bivariate approaches to ensure more reliable estimations. As shown in Fig. 1, maternal mortality rates remained a major global health concern in 2020, with significant country and regional disparities. Many low- and middle-income countries faced weak healthcare systems, limited access to skilled providers, inadequate infrastructure, and high fertility rates, all of which contributed to increased maternal mortality rates [2]. Additionally, high poverty levels and poor maternal nutrition further exacerbated maternal health risks. The COVID-19 pandemic worsened the already fragile healthcare systems in many African countries, leading to deteriorating maternal health outcomes [34]. In contrast, high-income countries—particularly in Europe, East Asia, and North America—maintained significantly lower maternal mortality rates in 2020.

Fig. 1.

Distribution of maternal mortality rates across countries

Figure 2 illustrates the global distribution of robotic flows (upper map) and stocks (lower map) in 2020. Countries without available data are omitted. The global AI flow showed significant progress, reflecting the growing integration of AI across healthcare, manufacturing, banking, and education. However, AI adoption varied widely across regions, influenced by technological infrastructure, investment levels, and digital preparedness. High-income countries, particularly in East Asia, North America, and Europe, recorded the highest AI inflows, driven by advanced digital infrastructure and substantial R&D investments. In East Asia alone, AI flow exceeded 150,000 units, with many countries expanding AI adoption in healthcare to enhance diagnostic accuracy, personalize treatment, and streamline clinical workflows [35]. In contrast, low-income countries, particularly in sub-Saharan Africa and South Asia, saw limited AI adoption in 2020. While some nations, such as South Africa and Kenya, have made strides in AI adoption—notably in agriculture and healthcare—overall AI flow in Africa remains low [36].

Fig. 2.

Maps of industrial and services robot flows and stocks

Countries with the highest AI stocks were predominantly high-income nations in North and Central America, Europe, and East Asia, where strong digital infrastructure and significant R&D investments drive AI expansion. In these regions, AI stocks are concentrated in healthcare, autonomous vehicles, defense, and finance, reflecting their strategic importance [37].

Difference-in-Difference estimations

The year 2000 marked a milestone in AI and robotics with the creation of Kismet, the first robot designed to simulate human emotions [38]. Two years later, in 2002, the release of the first Roomba, a consumer-grade autonomous robotic vacuum cleaner, showcased how robots could be integrated into everyday life for practical use. These innovations show rapid advancements in AI, making 2000 the logical post-treatment cut-off for assessing AI’s impact on maternal mortality.

The Difference-in-Differences (DiD) results in Table 3 and Fig. 3 reveal varying AI impacts based on development levels. In developing countries, the treated group had a higher MMR than the control group by 92.04 (113.82) per 100,000 live births for robot stock (flow). However, post-2000, MMR declined relative to pre-2000 levels, suggesting AI’s potential benefits. In developed countries, the treated group and post-2000 period showed lower maternal mortality rates, reinforcing AI’s role in improving maternal health outcomes.

Table 3.

The effects of AI on maternal mortality using difference in difference estimator

	(1)	(4)	(2)	(5)	(3)	(6)
	AI stock	AI Flow	AI stock	AI flow	AI stock	AI flow
Variables	All countries		Developed		Developing
treat	74.250c	86.238c	−12.265c	−13.138c	92.041c	113.826c
	(8.118)	(8.847)	(0.912)	(0.984)	(11.324)	(12.442)
post	−21.349b	−40.848c	−9.603c	−15.222c	−28.146b	−60.420c
	(9.712)	(7.709)	(0.974)	(0.771)	(14.174)	(11.386)
treat_post	−88.133c	−68.541c	4.8217c	11.002c	−109.832c	−77.365c
	(10.488)	(8.686)	(1.115)	(0.949)	(15.049)	(12.478)
Constant	203.884c	195.064c	26.433c	26.899c	268.800c	251.737c
	(6.762)	(7.242)	(0.684)	(0.7190)	(9.7455)	(10.556)
Observations	4,956	4,956	1,260	1,260	3,696	3,696
R-squared	0.140	0.141	0.338	0.326	0.175	0.176
Number of cid	177	177	45	45	132	132
ll	−30,011	−30,009	−4147	−4158	−22,845	−22,841
F	259.7	260.7	206.3	195.5	251.0	254.1

Standard errors in parentheses

^b, ^c denotes the level of significance at 5%, or 1%, respectively

Fig. 3.

Comparison of treated versus control groups in developed and developing countries

The interaction between treatment and post-treatment periods (treat_post), which captures AI’s actual impact, is significant and negative across all countries and in developing nations (−88.133 and −68.541 globally, −109.83 and −77.37 for developing countries for AI stocks and flows, respectively). This suggests that as AI technologies integrate into healthcare systems, their impact on reducing maternal mortality becomes more pronounced in developing countries over time. These findings align with previous research highlighting AI’s transformative role in improving healthcare access and outcomes in resource-limited settings [39]. AI adoption has a positive but less pronounced effect in developed countries, where maternal mortality rates are already low. The results indicate that greater AI adoption will benefit developing countries, where maternal mortality remains high, and AI implementation is still limited. This supports prior studies emphasizing AI’s long-term benefits in healthcare [9].

Panel ARDL modelling

Tables 4 and 5 present the panel ARDL model results assessing the impact of AI-driven robotics flow (aiapflow) on maternal mortality (mmr) across all countries, developed and developing countries. The results distinguish between short-run (SR), long-run (LR), and adjustment (ADJ) effects. In Table 6, the adjustment coefficient (ADJ term) for developing countries is significant at 5% and negative (−0.2717), indicating that 27% of deviations from the long-run maternal mortality equilibrium are corrected each period. In developed countries, the ADJ term is also significant and negative at both levels (−0.1336) and log (−0.0701), suggesting a 13.36% or 7.01% correction toward equilibrium per period. However, for all countries combined, the coefficient is positive (0.0488) and not statistically significant, implying weak or no global trend adjustment. These results indicate that maternal mortality in developing countries adjusts toward equilibrium slightly faster than in developed countries following a shock.

Table 4.

Panel ARDL modelling results for the impact of AI in industrial and services robotics flow on maternal mortality

	(1)	(2)	(3)	(4)	(5)	(6)
	All countries		Developing countries		Developed countries
Variables	Level	Log	Level	Log	Level	Log
ADJ
L.mmr	0.0268	0.0488	0.0347b	−0.2717b	−0.1336c	−0.0701b
	(0.0164)	(0.0328)	(0.0126)	(0.1263)	(0.0220)	(0.0272)
LR
L.aiapflow	−0.0289b	−0.7092c	−0.0386c	−0.1462c	0.0013a	0.0504
	(0.0125)	(0.2103)	(0.0107)	(0.0157)	(0.0007)	(0.4227)
SR
L2D.lmmr				0.3980	−0.0885	0.2909
				(0.2505)	(0.1483)	(0.2866)
L3D.lmmr				0.9369c		0.1254
				(0.2968)		(0.2455)
D.aiapflow	0.0008c	0.0346a	0.0021c	−0.0129	0.0001	−0.0051
	(0.0002)	(0.0182)	(0.0006)	(0.0154)	(0.0001)	(0.0267)
LD.aiapflow			−0.0015b	0.0114	−0.0003b	−0.0319
			(0.0006)	(0.0167)	(0.0001)	(0.0239)
L2D.aiapflow					−0.0001
					(0.0001)
L3D.aiapflow					−0.0001
					(0.0001)
Constant	−5.0879c	−0.4874a	−9.4008c	1.5030a	0.2323	0.1009
	(1.3263)	(0.2608)	(1.6618)	(0.7164)	(0.3554)	(0.2782)
Observations	27	27	26	24	24	24
R-squared	0.4531	0.1396	0.7290	0.6101	0.8885	0.6457

Source: Author’s computations

The dependent variable is maternal mortality ratio (per 100,000 live births) (mmr)

LR Represents the long-run results, SR Represents the short-run results, ADJ Represent the adjustment term

^a, ^b, ^c denotes the level of significance at 10%, 5%, or 1%, respectively. Standard errors in parentheses

Table 5.

Diagnostic statistics test results for (AI) in robotics flow and maternal mortality model

	All countries		developing		Developed
	non-logged	logged	non-logged	logged	non-logged	logged
Durbin-Watson (DW)	2.288	2.102	2.032	2.218	1.448	1.782
BG Chi2	0.943	1.952	0.022	2.154	1.362	0.653
BG P > Chi2	0.331	0.162	0.882	0.142	0.243	0.419
White Chi2	19.15	22.84	14.52	24.00	9.99	5.97
White P > Chi2	0.512	0.297	0.803	0.404	0.351	0.310
PSS F	9.943c	1.947c	19.222c	6.486c	25.023c	5.876c
PSS p-val I(0)	0.004	0.431	0.000	0.030	0.000	0.042
PSS p-val I(1)	0.008	0.555	0.000	0.057	0.000	0.076

Source: Authors’ computations

^c denotes the level of significance at 1% respectively

Table 6.

Panel ARDL modelling results for the impact of artificial intelligent (AI) in industrial and services robotics stock on maternal mortality

	(1)	(2)	(3)	(4)	(5)	(6)
	all countries		Developing		Developed
Variables	level	log	level	log	level	log
ADJ
L.mmr	0.0387b	0.1381c	0.0337b	−0.1395	−0.0508 b	0.0383
	(0.0159)	(0.0390)	(0.0121)	(0.1753)	(0.0227)	(0.0404)
LR
L.aiapstock	−0.0042 c	−0.8390 c	−0.0108 a	−0.1093	0.0028	−4.8868
	(0.0011)	(0.0718)	(0.0056)	(0.0956)	(0.0017)	(3.7848)
SR
LD.lmmr				0.0146	−0.6682 c	−0.4546 a
				(0.2707)	(0.1298)	(0.2301)
L2D.lmmr				0.3786	−0.3283 b	0.1430
				(0.2555)	(0.1122)	(0.2320)
L3D.lmmr				0.7905b		0.3188
				(0.3016)		(0.2096)
D.aiapstock	0.0002c	0.1158c	0.0018c	−0.0081	0.0000	−0.1528
	(0.0000)	(0.0296)	(0.0006)	(0.0721)	(0.0001)	(0.2058)
LD.aiapstock			−0.0023 c	−0.1331 a	−0.0003 c	−0.5995 b
			(0.0008)	(0.0735)	(0.0001)	(0.2433)
L2D.aiapstock					−0.0002 b
					(0.0001)
L3D.aiapstock					−0.0002 b
					(0.0001)
Constant	−6.7561 c	−1.6960 c	−9.2775 c	0.7979	−3.3681 c	−1.9870 b
	(1.4141)	(0.4350)	(1.6000)	(1.0579)	(0.7907)	(0.6855)
Observations	27	27	26	24	24	24
R-squared	0.5465	0.3959	0.7522	0.6218	0.9507	0.7686

Source: Author’s computations

The dependent variable is maternal mortality ratio (per 100,000 live births) (mmr)

LR Represents the long-run results, SR Represents the short-run results, ADJ Represent the adjustment term

^a, ^b, ^c denotes the level of significance at 10%, 5%, or 1%, respectively. Standard errors in parentheses

Figure 4, depicting the cumulative sum (CUSUM), shows that all variables remain within the 95% confidence band, confirming model stability over time. This is further supported by diagnostic test results in Table 7, indicating that the estimated models are statistically reliable. The Durbin-Watson (DW) statistic suggests minimal autocorrelation, with most values close to 2 (e.g., 2.288, 2.102, 2.218). The Breusch-Godfrey (BG) Chi2 and p-values confirm no significant serial correlation in any model. White’s test results indicate no significant heteroskedasticity, suggesting consistent error variance. Finally, the Pesaran, Shin, and Smith (PSS F-statistics) and p-values confirm cointegration in most models, reinforcing the existence of a stable long-term relationship between AI in robotics and maternal mortality across different countries.

Fig. 4.

CUSUM graphs for AI Robotic flows

Table 7.

Diagnostic statistics test results for (AI) in robotics stock and maternal mortality model

	All countries		developing		Developed
	non-logged	logged	non-logged	logged	non-logged	logged
DW	2.070	2.082	2.058	2.324	1.933	2.080
BG Chi2	0.189	0.919	0.043	3.091	0.031	0.616
BG P > Chi2	0.663	0.338	0.836	0.079	0.860	0.432
White Chi2	9.37a	18.89c	9.35	24.00	24.00	24.00
White P > Chi2	0.095	0.002	0.808	0.404	0.404	0.404
PSS F	14.463c	7.863b	9.102b	7.037b	64.961c	16.453c
PSS p-val I(0)	0.000	0.012	0.007	0.023	0.000	0.001
PSS p-val I(1)	0.001	0.023	0.014	0.044	0.000	0.001

Source: Author’s computations

^a, ^b, ^c denotes the level of significance at 10%, 5%, or 1%, respectively

The long-run (LR) effects of AI Robotics Flow indicate a significant negative relationship with maternal mortality across all countries, suggesting that increased AI adoption reduces maternal mortality over time. In developing countries, the effect is stronger and more significant, highlighting a greater long-term impact of AI robotics in reducing maternal mortality. Conversely, in developed countries, the coefficient is positive but insignificant, indicating no strong long-run effect of AI robotics flow on maternal mortality. The negative long-run coefficients in developing countries suggest that AI-driven technologies, particularly in industrial and service robotics, enhance healthcare efficiency and accessibility, ultimately reducing maternal mortality. This aligns with technological diffusion theories, where advanced technologies improve healthcare service delivery, reduce human error, and strengthen maternal health monitoring. The stronger effect in developing countries may reflect a"technological catch-up"process, where adopting AI-driven innovations enhances healthcare infrastructure, improves medical access, and supports maternal health programs [14, 40, 41].

The positive short-run effects observed globally and in developing countries suggest transitional disruptions as AI integration initially strains healthcare systems due to high costs, technical challenges, or workforce displacement before long-term benefits emerge [6]. In developed countries, the absence of a significant long-run impact may indicate AI saturation, where additional investments yield diminishing returns on maternal health outcomes due to already advanced healthcare infrastructure. The relationship between AI and maternal health aligns with key economic and public health theories. The innovation and public health theory suggests that technological advancements enhance efficiency and accessibility, improving health outcomes [42]. Meanwhile, the diffusion of innovations theory explains varying impacts across countries, emphasizing how technological adoption stages and contextual factors influence AI’s effectiveness [43].

Panel ARDL modelling results and discussion for the impact of artificial intelligence (AI) in robotics stock on maternal mortality

Tables 6 and 7 present the panel ARDL (Auto-Regressive Distributed Lag) model results, assessing the impact of AI-driven robotics stock (aiapstock) on maternal mortality across global, developed, and developing countries. Table 8 shows that the adjustment coefficients for maternal mortality are positive and significant, indicating a degree of inertia in maternal mortality rates. This suggests deviations from the long-run equilibrium due to AI robotics stock changes are corrected over time, with a faster adjustment observed in all countries and developing nations. Figure 5, displaying the cumulative sum, reveals that some variables fall outside the 95% confidence band in global and developed country samples. Despite this, the model remains reliable for interpretation. This is further supported by diagnostic test results in Table 8, which confirm that the models are generally well-specified, though minor heteroscedasticity issues exist in some cases. The Durbin-Watson and Breusch-Godfrey tests indicate no significant autocorrelation, ensuring the model’s standard errors remain unbiased. However, White’s test detects heteroscedasticity, particularly in logged models for all countries and developed countries. Importantly, significant PSS F-statistics across all models confirm a long-run relationship between maternal mortality and AI robotics stock, suggesting cointegration. This indicates that maternal mortality and AI advancements move together over time, with deviations from equilibrium corrected in the future. The findings reinforce that AI has long-term implications for maternal mortality reduction across different country groups.

Table 8.

Fixed effect regression results (forecast)

	(1)	(2)	(3)	(4)	(5)	(6)
	All countries		Developing		Developed
Variables	AI stock	AI flow	AI stock	AI flow	AI stock	AI flow
AI stock	−0.00003 b		−0.00005 a		0.000023
	(0.00001)		(0.00002)		(0.000017)
AI flow		−0.00013 b		−0.000152 a		0.000049
		(0.00006)		(0.000088)		(0.000096)
ha_ind	−5.48207 c	−5.52103 c	−5.478213 c	−5.567772 c	−4.370444 c	−4.189829 c
	(0.92313)	(0.91457)	(2.10145)	(2.098164)	(0.843351)	(0.841969)
oca_bsw	−2.66166 c	−2.65660 c	−2.78832 c	−2.790786 c	0.457653	0.406633
	(0.10833)	(0.10832)	(0.15677)	(0.156862)	(0.352817)	(0.351136)
cm_ind	−1.64990	−1.65288	−0.93356	−1.034104	−4.724388 a	−4.706678 a
	(2.22262)	(2.22021)	(3.98890)	(3.989495)	(2.541885)	(2.546307)
Cons	287.33172c	286.66343c	314.65147c	314.73471c	−35.271658	−30.051272
	(10.52861)	(10.52733)	(14.80255)	(14.81535)	(35.458956)	(35.279466)
Obs	709	709	233	233	476	476
R-sq	0.52	0.52	0.64	0.64	0.078	0.08
Num. of cid	62	62	28	28	34	34
Log Lik	−2346	−2345	−842.5	−842.7	−1404	−1405

Source: Author’s computations

^a, ^b, ^c denotes the level of significance at 10%, 5%, or 1%, respectively

Fig. 5.

CUSUM graphs for AI Robotic stock

At the global level, AI robotics stock has a statistically significant negative long-run impact on maternal mortality, indicating that expanding AI technologies in industry and services reduces maternal mortality worldwide. A 1-unit increase in AI robotics stock (aiapstock) lowers MMR by 0.0042 units (level) or 0.8390% (log). AI also significantly reduces maternal mortality in the short run, particularly in the log model (0.1158), underscoring its immediate benefits. AI-driven diagnostic tools, automated surgical robots, and improved healthcare delivery contribute to better maternal health outcomes by enhancing obstetric care and addressing complications more effectively.

The long-run impact is mixed for developing countries (see Columns 3 and 4). In the level model, AI robotics stock significantly reduces maternal mortality (0.0108). However, the log model shows no significance, suggesting that while AI stock influences maternal mortality, its effects may not scale logarithmically. In the short run, AI exhibits significant adverse effects at first and second lags, indicating immediate and sustained reductions in maternal mortality. The impact is more pronounced in developing countries, where healthcare access and quality gaps are larger. AI technologies help bridge these gaps by enhancing diagnostics, expanding telemedicine, and improving healthcare efficiency, ultimately reducing maternal mortality rates. However, the insignificance of the log model suggests that AI penetration remains too limited to generate large-scale effects. These findings align with the health production function framework, which views health outcomes as a function of technological inputs [44]. AI robotics enhances healthcare efficiency and service delivery, reinforcing prior studies highlighting AI’s positive role in reducing mortality in developing regions [8, 45–47].

For developed countries (see Columns 5 and 6), the long-run impact of AI robotics stock on maternal mortality is insignificant, both in the level and log models, suggesting that AI does not significantly influence maternal mortality over time. This is likely due to already high healthcare standards, where additional AI investments yield diminishing returns. In the short run, several lagged AI robotics stock terms (LD.aiapstock, L2D.aiapstock, L3D.aiapstock) show significant negative coefficients, indicating that AI has a more immediate impact. This suggests cutting-edge AI technologies enhance short-term medical interventions, improving operational efficiency, resource allocation, and emergency maternal healthcare responses. The contrast between developed and developing countries aligns with economic convergence theory, which posits that less developed nations experience greater benefits from adopting advanced technologies [48, 49]. Since developing countries start from a lower healthcare baseline, AI integration has a more pronounced effect on maternal mortality reduction than developed nations.

Fixed effect regression results and discussion for the impact of artificial intelligence (AI) on maternal mortality

For all countries (Columns 1 and 2, Table 8), the AI stock coefficient (−0.00003) is statistically significant at the 5 percent level, indicating that higher AI stock marginally reduces maternal mortality rates. This suggests that accumulated industrial and service robotics improve maternal health outcomes. The AI flow coefficient (−0.00013) is also significant at the 5 percent level, showing a stronger negative impact on maternal mortality rates than AI stock. This implies that advancements in AI, particularly in healthcare services, are more effective in reducing maternal deaths, aligning with previous studies [50–52].

For developing countries (Columns 3 and 4), the AI stock coefficient (−0.00005) is significant at the 10 percent level, reinforcing that increased AI stock helps lower maternal mortality, with a slightly larger effect than in the overall sample. AI applications in maternal health services, such as robotic-assisted procedures and diagnostics, can help bridge healthcare infrastructure gaps. The AI flow coefficient (−0.000152) is also significant at the 10 percent level, indicating that new AI technologies more effectively reduce maternal mortality in developing countries by addressing healthcare deficiencies through cost-effective diagnostics and telemedicine solutions. The significant negative coefficients for AI stock and flow suggest that AI technologies can fill critical gaps in maternal healthcare, particularly where access to skilled medical personnel is limited. These findings align with technological leapfrogging theory, which posits that developing nations benefit more from newer technologies than developed countries with established infrastructures [53, 54]. For developed countries (Columns 5 and 6), the AI stock coefficient (0.000023) and AI flow coefficient (0.000049) are positive but not significant, indicating no clear relationship between AI adoption and maternal mortality rates. This suggests that AI may enhance efficiency but does not significantly impact maternal health outcomes in countries with already advanced healthcare systems [55].

The global stock of robots tripled to 2.25 million in 2018 [24], with projections estimating 20 million robots by 2030, representing a 19% annual growth rate. This rate is used to extend yearly flows and stocks of industrial and service robots to 2035, though it remains conservative given the 48% increase in professional service robot sales in 2022 [56].

To analyze the impact of AI-driven robotics (AI stock and AI flow) on maternal mortality(MMR, panel fixed-effects models are estimated using actual data from 1993 to 2020 across global, developing, and developed countries. Figure 6 presents forecasts based on fixed-effects (FE) regression models, controlling for health access, obstetric care, and comorbidities while allowing AI indicators (stock and flow) to vary. The forecasts remain reliable, as indicated by the narrow 95% confidence intervals (shaded grey area), particularly in developed countries.

Fig. 6.

Forecast graphs of the impact of AI robotics on maternal mortality

The forecast suggests a slightly more substantial impact of AI flow on maternal mortality (MMR) than AI stock. Globally, MMR follows a clear downward trend in both AI stock and flow models. The actual values (blue line) remained stable from 1990 to 2020. The forecast (red line) projects a significant decline from 2020 onward, with MMR dropping below 20 per 100,000 live births by 2035. The forecast shows a steeper initial decline in developing countries, with MMR falling below 50 by 2030 in both models. However, the decline is more pronounced in the AI flow model compared to AI stock. Furthermore, the forecast predicts a modest further reduction in developed countries, with low MMR (below 20 per 100,000). The AI stock model shows little change beyond 2020, while the AI flow model indicates a slight decline, keeping MMR below 10.

Conclusion

This study highlights the transformative potential of AI in reducing maternal mortality, particularly in achieving Sustainable Development Goal 3.1. Using Difference-in-Differences (DiD), Dynamic Panel ARDL, and fixed effects regression, the study examines AI’s impact on maternal health across 70 countries, distinguishing between developed and developing regions.

The findings reveal that AI, particularly in industrial and service robotics, significantly reduces maternal mortality globally, with more substantial effects in developing countries, where healthcare infrastructure is weaker and maternal mortality is very high. This underscores the critical role of AI in bridging healthcare gaps and improving access to maternal health services. In developed countries, where maternal mortality rates are already low, AI adoption continues to enhance healthcare outcomes, though its impact is less pronounced. This is attributed to well-established healthcare systems, where AI’s marginal benefits are smaller than developing regions.

Key policy recommendations

Governments in developing countries should prioritize AI investments in maternal healthcare to reduce high maternal mortality rates. This includes promoting AI-powered diagnostics, telemedicine, and AI-assisted healthcare training to improve outcomes in resource-constrained settings.

Countries must strengthen digital infrastructure to maximize AI’s benefits, including reliable internet access, digital literacy programs, and AI research and development. This is essential for effective AI integration into healthcare systems, particularly in developing nations.

Policymakers should leverage AI to bypass traditional healthcare barriers, enabling faster adoption of advanced diagnostic and treatment tools. This can bridge critical healthcare access and quality gaps, especially in rural and underserved areas.

Governments and international organizations must ensure that AI adoption does not widen healthcare inequalities. Policies should promote affordable, accessible AI tools, particularly for low-income and rural populations.

Caveats and areas for further research

While this study provides valuable insights into AI’s role in reducing maternal mortality, some challenges must be acknowledged. Biases in AI data may lead to selection effects, as current datasets focus on industrial robotics rather than specific maternal health applications. Healthcare system disparities further complicate AI adoption, with infrastructure, workforce capacity, and policy environments influencing outcomes differently across countries. Additionally, high investment requirements in developing countries limit AI integration, while privacy concerns raise ethical issues regarding data security and patient confidentiality. The dynamic nature of AI data means that available datasets may not capture all dimensions of maternal health, potentially impacting long-term applicability. Moreover, ethical challenges, including algorithmic bias and regulatory inconsistencies, must be addressed to ensure equitable AI deployment in maternal healthcare.

Despite these limitations, this study provides a critical foundation for understanding AI’s transformative potential in reducing maternal mortality. Future research should prioritize health-specific AI datasets, machine learning-based causal inference models, and longitudinal analyses to refine AI’s role in maternal healthcare. Addressing these challenges through policy frameworks, digital infrastructure investments, and ethical AI governance will be essential to ensuring AI’s sustainable impact on global maternal health.

Appendix 1

Table 9.

Variable description and data sources

Indicator	Computation	Constituent variables	Sources
mmr_nw	Maternal mortality ratio (per 100,000 live births)	mmr	WHO database
Artificial intelligent indicators
aiapflow	Industrial and services robotics flow	aiapflow	World Robotics Congress database
aiapstock	Industrial and services robotics stock	aiapstock	World Robotics Congress database
Ai_ind	Index of artificial intelligence, from the Principal Component analysis (PCA) of:
	Digitally deliverable services exports, Current USD (millions)	ddsx_cusd	UNCTAD database
	Digitally deliverable services imports, Current USD (millions)	ddsm_cusd	UNCTAD database
	Frontier technology readiness index, R&D	ftri_rd	UNCTAD database
	Frontier technology readiness index, overall	ftri_oi	UNCTAD
	Frontier technology readiness index, industry activity	ftri_ia	UNCTAD database
	Frontier technology readiness index, ICT	ftri_ict	UNCTAD database
	Frontier technology readiness index, access to finance	ftri_af	UNCTAD database
	ICT services exports Current USD (millions)	ictstx_cusd	UNCTAD database
	ICT services exports percentage of total world	ictstx_petw	UNCTAD database
	ICT services imports Current USD (millions)	ictstm_cusd	UNCTAD database
	Share of ICT goods re-imports	ictg_rm	UNCTAD database
	Share of ICT goods exported	ictg_x	UNCTAD database
	Share of ICT goods imported	ictg_m	UNCTAD database
cm_ind	Index of comorbidities, as a Principal Component analysis (PCA) of:
	Prevalence of anemia among women of reproductive age (% of women ages 15–49)	cm_apw	WHO, GBD database
	Suicide mortality rate, female (per 100,000 female population)	cm_srf	WHO, GBD database
	Women’s share of population ages 15 + living with HIV (%)	cm_hivpf	WHO, GBD database
ha_ind	Index of health access, from the Principal Component analysis (PCA) of:
	Nurses and midwives (per 1,000 people)	ha_nmt	WDI database
	Physicians (per 1,000 people)	ha_ppt	WDI database
he_ind1	Index of health expenditure computed via Principal Component analysis (PCA) of:
	Current health expenditure (% of GDP)	he_gdp	WDI database
	Current health expenditure per capita (current US$)	he_pc	WDI database
	Current health expenditure per capita, PPP (current international $)	he_ppp	WDI database
he_oops	Out-of-pocket expenditure (% of current health expenditure)	he_oops	WDI database
inf_ind	Index of infrastructure, from the Principal Component analysis (PCA) of:
	Access to electricity (% of population)	inf_ael	WDI database
	Individuals using the Internet (% of population)	inf_inp	WDI database
se_indp	Socioeconomic indicator correlating positively with mmr, from the PCA of:
	GDP per capita (constant 2015 US$)	se_gdppc	WDI database
	People with basic handwashing facilities including soap and water (% of population)	se_hwf	WDI database
	Employment to population ratio, 15 +, female (%) (national estimate)	se_eprf	WDI database
	Labor force participation rate, female (% of female population ages 15–64) (modeled ILO estimate)	se_flfpr	WDI database
se_indn	Socioeconomic indicator correlating negatively with mmr, from the PCA of:
	Poverty headcount ratio at $2.15 a day (2017 PPP) (% of population)	se_phr2_15	WDI database
	Poverty headcount ratio at $3.65 a day (2017 PPP) (% of population)	se_phr3_65	WDI database
	Poverty headcount ratio at $6.85 a day (2017 PPP) (% of pop.)	se_phr6_85	WDI database
	Prevalence of undernourishment (% of population)	se_unp	WDI database
	Age dependency ratio (% of working-age population)	se_adr	WDI database
oca_ind	Proxy of Obstetric Care Availability, measured by:
	Births attended by skilled health staff (% of total)	oca_bsw	WDI database
de_frt	Fertility rate, total (births per woman)	de_frt	WDI database

Source: Author’s compilation

Appendix 2

Table 10.

AI robots usage country list by levels of development

Developing	Developed
Argentina	Australia
Brazil	Austria
Chile	Belarus
China	Belgium
Colombia	Bosnia and Herzegovin
Costa Rica	Bulgaria
Denmark	Canada
Egypt	Czechia
India	Estonia
Indonesia	Finland
Iran (Islamic Republi	France
Kuwait	Germany
Malaysia	Greece
Mexico	Hungary
Morocco	Iceland
Oman	Ireland
Pakistan	Israel
Peru	Italy
Philippines	Japan
Puerto Rico	Latvia
Qatar	Lithuania
Saudi Arabia	Malta
Sierra Leone	Netherlands (Kingdom
Singapore	New Zealand
South Africa	Norway
Thailand	Poland
Tunisia	Portugal
Türkiye	Republic of Korea
United Arab Emirates	Republic of Moldova
Uzbekistan	Romania
Venezuela (Bolivarian	Russian Federation
Viet Nam	Serbia
	Slovakia
	Spain
	Sweden
	Switzerland
	Ukraine
	United Kingdom of Gre
	United States of Amer

Source: Author’s compilation

Authors'contributions

N. Ngepah conceived the key ideas for this research paper. He collected and analysed the data. N. Ngepah, Charles S. Saba, AEN Mouteyica, and A. Ohonba wrote the introduction.  N. Ngepah, Charles S. Saba, AEN Mouteyica, and A. Ohonba wrote the literature review. N. Ngepah and Charles S. Saba wrote the methodology. N. Ngepah, Charles S. Saba, and AEN Mouteyica wrote results, and conclusion.

Funding

The Future of Life Institute.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

This article does not contain any studies with human participants or animals performed by the author. Ethical approval for this type of study is not required by our institution.

Consent for publication

This manuscript is an original work the author(s) produced. NN, CSS, AENM, and AO know the content and approve its submission. It is also important to mention that the manuscript has not been published elsewhere in part or entirety and is not being considered by another journal. The author(s) has consented to submit this article for publication in Social Science & Medicine.

Competing interests

The authors declare no competing interests.