Unsafe drinking water and human health: A global umbrella review of disease outcomes, intervention effectiveness, and policy implications

Summary

Access to safe drinking water is a fundamental determinant of global health. In this umbrella review, we synthesized evidence from 25 systematic reviews and meta-analyses, covering 158 outcomes, to assess health risks and intervention effectiveness (PROSPERO: CRD420251001778). Employing a rigorous methodological framework including A Measurement Tool to Assess Systematic Reviews 2 (AMSTAR 2); Evidence Classification; the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE); and simplified evidence-to-decision criteria, we identified significant associations between unsafe drinking water and various health conditions, including infectious diseases, cancers, cardiovascular diseases, and adverse maternal-child health outcomes. Importantly, while the certainty of evidence for precise risk attribution remains limited, evidence supporting effective interventions is robust. Point-of-use (POU) filtration reduces childhood diarrhea by 52% (relative risk [RR] = 0.48; moderate-certainty evidence), and defluoridation effectively prevents fluorosis. Overall, these findings support a shift in policy focus: despite uncertainties in exact risk quantification, public health strategies should prioritize immediate implementation of proven interventions to safeguard vulnerable populations.

Graphical abstract

Highlights

• •Unsafe water is linked to cancer, infection, cardiovascular, and maternal-neonatal risks
• •Water filtration and defluoridation effectively reduce disease risks
• •Risk attribution robustness varies; intervention benefits are robust
• •Policy should prioritize proven solutions over precise risk estimates

Li et al. synthesize global evidence on risks from unsafe drinking water and related interventions. They find variable statistical robustness in attributing health risks; however, interventions such as filtration consistently demonstrate clear benefits. The results highlight the need to shift policies toward implementing proven water safety solutions.

Introduction

Access to safe drinking water is fundamental to human health. According to the Guidelines for Drinking-water Quality (fourth edition, first addendum), safe drinking water is defined as water that “does not represent any significant risk to health over a lifetime of consumption, including different sensitivities that may occur between life stages.”¹ However, approximately 2.1 billion people still lacked access to safely managed drinking-water services in 2024, mainly in low-income countries with pronounced urban-rural disparities.² This exposes populations to waterborne infectious diseases, such as diarrhea and typhoid, as well as chronic health risks from chemical contaminants, including arsenic, fluoride, and lead.^3,4 Consequently, ensuring drinking-water safety is a central priority of the UN Sustainable Development Goal 6.

However, due to insufficient monitoring, multiple pollution sources, and diverse disparities, a comprehensive understanding of the health and intervention outcomes related to unsafe drinking water remains limited.^5,6 Although many meta-analyses link unsafe drinking water to adverse health outcomes, most focus on single pollutants, outcomes, or interventions. Few integrate multiple contamination sources and chronic health effects or systematically compare intervention effectiveness. Moreover, some existing umbrella reviews have preliminarily synthesized water pollution research, but they neither accurately reflect health risks specifically related to drinking water nor systematically integrate diverse pollutants, multiple health outcomes, and interventions into an evidence-to-decision (EtD) framework to clearly guide policy decisions. For instance, Gao’s review exclusively addressed neurodevelopmental impacts without considering broader health outcomes or interventions.⁷ Thomson’s work primarily examined the influence of public health policies on inequalities in high-income countries, mentioning drinking-water safety only briefly, thereby limiting global applicability.⁸ Additionally, some studies on environmental risk factors either excluded drinking-water contamination or did not specify if contamination originated from drinking water.⁹ This lack of relevant, precise evidence significantly complicates the direct assessment of drinking-water safety. Consequently, conducting a more comprehensive umbrella review on health outcomes and intervention effectiveness related to unsafe drinking water is essential to address these gaps and support evidence-based policy development.

In this study, we synthesize evidence on diverse pollutants, multiple health outcomes, and drinking-water interventions into an integrated, evidence-based framework to systematically assess the risks and effectiveness of interventions related to unsafe drinking water. To enhance methodological rigor, we apply the Evidence Classification framework and A Measurement Tool to Assess Systematic Reviews 2 (AMSTAR 2) tool to evaluate the quality of included meta-analyses and assess clinical credibility of the evidence using the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) framework. Additionally, we adopt a simplified EtD framework to help translate evidence into potential policy recommendations. This study aims to provide a comprehensive global assessment of the health impacts and intervention effectiveness related to unsafe drinking water, offering evidence-based insights to potentially inform targeted interventions and contribute to global progress toward universal access to safe drinking water.

Results

Figure 1 illustrates the literature search and selection process. Ultimately, 25 systematic reviews and meta-analyses covering 158 outcomes were included and re-analyzed (Data S1 and S2).

Figure 1.

Study selection

Flowchart illustrating the literature search and selection process.

Association between unsafe drinking water and infectious diseases

Several systematic reviews and meta-analyses have assessed the associations between microbiologically contaminated drinking water and the risk of infectious diseases such as diarrhea and typhoid fever; however, the overall evidence quality remains low (Figure 2). Regarding diarrhea, one meta-analysis reported a significantly increased risk of diarrhea among populations exposed to drinking water contaminated specifically with Escherichia coli (EC), showing a relative risk (RR) of 1.54 (95% confidence interval [CI]: 1.37–1.74; Class IV: weak evidence, GRADE: very low). Another broader meta-analysis, encompassing exposure to either EC or thermotolerant coliforms, indicated a slight increase in diarrhea risk (RR = 1.25, 95% CI: 0.99–1.58), but this did not reach statistical significance. Similarly, when evaluating exposure to thermotolerant coliforms alone, no significant association with diarrhea risk was observed (RR = 1.07, 95% CI: 0.79–1.46).

Figure 2.

Associations between unsafe drinking water and health risks (infectious disease, cancer, and maternal and child health)

Summary results from re-analysis of included meta-analyses using random-effects models. Evidence classes range from Class I (strong evidence) to Class IV (weak evidence). GRADE ratings assess evidence quality from high to very low, considering methodological limitations and consistency.

For typhoid fever, a meta-analysis showed a significantly increased risk among populations exposed to microbiologically contaminated drinking water (OR = 2.78, 95% CI: 1.80–4.32; Class III: suggestive evidence, GRADE: very low).

Association between unsafe drinking water and cancer risk

The overall evidence suggests limited but clear associations between specific drinking-water contaminants and cancer risks, although most evidence remains at weak or suggestive levels with generally low GRADE ratings (Figure 2). Regarding nitrate contamination, several meta-analyses indicated slight increases in cancer risk associated with high nitrate exposure, particularly for colon cancer (odds ratio [OR] = 1.15, 95% CI: 1.04–1.28), gastric cancer (OR = 1.16, 95% CI: 1.02–1.33), and glioma (OR = 1.16, 95% CI: 1.08–1.26). However, these findings represent Class IV (weak evidence) with very low GRADE ratings. For rectal cancer (OR = 1.11, 95% CI: 0.95–1.31) and pancreatic cancer (OR = 1.09, 95% CI: 1.00–1.19), the effect estimates approached statistical significance but were not clearly significant. Additionally, nitrate exposure showed no statistically significant associations with bladder cancer (OR = 1.00), esophageal cancer (OR = 1.06), non-Hodgkin lymphoma (OR = 1.07), breast cancer (OR = 1.07), kidney cancer (OR = 1.22), or ovarian and uterine corpus cancers (OR = 1.49), all rated as very low GRADE evidence. Exposure to trihalomethanes (THMs) showed statistically significant positive associations with increased risks of rectal cancer (RR = 1.15, 95% CI: 1.07–1.24; Class III: suggestive evidence, GRADE: low; n = 90,195) and bladder cancer (RR = 1.33, 95% CI: 1.04–1.71; Class IV: weak evidence, GRADE: very low; n = 83,360).

Inorganic arsenic exposure through drinking water demonstrated a stronger and clearer association with liver cancer mortality. A meta-analysis found that high exposure to arsenic-contaminated water significantly increased liver cancer mortality risk by approximately 80% (standardized mortality ratio [SMR] = 1.80, 95% CI: 1.53–2.10, Class II: highly suggestive evidence, GRADE: very low; n = 2,252). Gender subgroup analyses revealed an even more pronounced association with males (n = 1,449; SMR = 1.94, 95% CI: 1.49–2.53; Class III: suggestive evidence, GRADE: very low) compared with females (n = 788; SMR = 1.69, 95% CI: 1.26–2.28; Class III: suggestive evidence, GRADE: very low).

Association between unsafe drinking water and maternal, neonatal, and child diseases

The overall evidence regarding the association between drinking-water contamination and maternal, neonatal, and child health outcomes is clear but limited (Figure 2). Arsenic-contaminated drinking water was relatively clearly associated with adverse pregnancy outcomes. Specifically, high arsenic exposure was significantly associated with increased risks of stillbirth (OR = 1.75, 95% CI: 1.33–2.31), neonatal mortality (OR = 1.51, 95% CI: 1.26–1.82), infant mortality (OR = 1.28, 95% CI: 1.12–1.45), and spontaneous abortion (OR = 1.99, 95% CI: 1.29–3.07). The evidence strength for these associations ranged mainly from Class III to IV (suggestive to weak evidence), with very low GRADE ratings. High arsenic exposure was also associated with reduced birth weight (average reduction of 50.53 g, 95% CI: −88.64 to −12.41; Class IV: weak evidence, GRADE: low). However, the association between arsenic exposure and preterm birth (OR = 1.41, 95% CI: 0.80–2.49) was not statistically significant.

Exposure to chlorinated disinfection by-products showed significant associations with congenital anomalies (OR = 1.18, 95% CI: 1.01–1.37; Class IV: weak evidence) and ventricular septal defects (OR = 1.58, 95% CI: 1.21–2.07; Class III: suggestive evidence). Exposure to THMs during pregnancy was significantly associated with an increased overall risk of major birth defects (OR = 1.46, 95% CI: 1.16–1.83; Class IV: weak evidence). All the above outcomes had very low GRADE ratings. Additionally, children in fluoride-contaminated regions exhibited significantly lower intelligence quotient (IQ) scores compared with non-exposed children (standardized mean difference [SMD] = −10.98, 95% CI: −13.82 to −8.14; Class II: highly suggestive, GRADE: very low). Furthermore, comprehensive drinking-water contamination was weakly associated with maternal mortality risk (OR = 1.75, 95% CI: 1.22–2.53), classified as Class IV: weak evidence with a very low GRADE rating.

Association between unsafe drinking water and cardiovascular diseases

Multiple meta-analyses have evaluated the association between arsenic exposure in drinking water and cardiovascular disease risk; however, the evidence strength and consistency of findings vary (Figure 3). At relatively low exposure levels (10 μg/L vs. 1 μg/L), stronger statistical evidence comes from a meta-analysis that demonstrated a significant 41% increase in fatal and non-fatal cardiovascular disease risk (RR = 1.41, 95% CI: 1.25–1.60; Class I: convincing evidence, GRADE: very low; n = 5,053). In addition, a meta-analysis found a significant association between arsenic exposure and hypertension risk (RR = 1.93, 95% CI: 1.40–2.67; Class I: convincing evidence; n = 1,505). Another large-scale meta-analysis indicated a significant increase in coronary heart disease mortality associated with arsenic exposure (RR = 1.50, 95% CI: 1.15–1.95; Class IV: weak evidence; n = 182,651). Additionally, the overall risk of coronary heart disease was significantly elevated (RR = 1.40, 95% CI: 1.18–1.67; Class III: suggestive evidence; n = 6,181). Both of these outcomes had very low GRADE ratings.

Figure 3.

Associations between unsafe drinking water and cardiovascular and other health risks

Summary results from re-analysis of included meta-analyses using random-effects models. Evidence classes range from Class I (strong evidence) to Class IV (weak evidence). GRADE ratings assess evidence quality from high to very low, reflecting methodological quality and consistency.

Regarding outcomes showing slightly elevated risks, we revealed a weak but significant association between arsenic exposure and overall cardiovascular mortality (RR = 1.17, 95% CI: 1.05–1.32; Class IV: weak evidence, GRADE: very low; n = 321,830). Another outcome demonstrated a mild but significant positive association with carotid atherosclerosis risk (RR = 1.23, 95% CI: 1.04–1.46; Class IV: weak evidence, GRADE: very low; n = 19,032). However, fatal and non-fatal stroke risk (RR = 1.11, 95% CI: 0.98–1.26; n = 61,439) and stroke mortality (RR = 1.13, 95% CI: 0.79–1.63; n = 242,004) did not reach statistical significance, both outcomes receiving very low GRADE ratings.

At relatively higher arsenic exposure levels (20 μg/L vs. 10 μg/L), a meta-analysis identified a small but significant increase in coronary heart disease mortality risk (RR = 1.16, 95% CI: 1.07–1.25; Class III: suggestive evidence; n = 330,744), although the GRADE rating remained very low. Additionally, a meta-analysis found a modest increase in overall cardiovascular disease incidence risk (RR = 1.09, 95% CI: 1.04–1.14; Class III: suggestive evidence; n = 6,131). In contrast, stroke mortality (RR = 1.06, 95% CI: 0.94–1.19; n = 330,744) and stroke incidence (RR = 1.08, 95% CI: 1.00–1.18) at the same exposure level showed no statistically significant associations, both with very low GRADE ratings.

Association between unsafe drinking water and other health risks

Several systematic reviews have evaluated associations between arsenic exposure from drinking water and urinary arsenic metabolism markers as well as pulmonary function indicators (Figure 3). However, the overall quality of evidence remains relatively low, predominantly classified as weak (Class IV), with very low GRADE ratings.

Specifically, a meta-analysis found that urinary dimethylarsinic acid (DMA) percentage was significantly lower in arsenic-exposed populations compared with non-exposed groups (SMD = −0.58, 95% CI: −0.97 to −0.19; Class IV: weak evidence; n = 3,763). Similarly, another analysis indicated significant reductions in both the primary methylation index (SMD = −0.59, 95% CI: −1.07 to −0.11) and secondary methylation index (SMD = −0.29, 95% CI: −0.52 to −0.05), both classified as Class IV weak evidence (n = 2,982). Conversely, urinary monomethyl arsenic percentage was significantly elevated in the exposed group (SMD = 0.52, 95% CI: 0.06 to 0.98; Class IV: weak evidence; n = 3,783). Although other urinary arsenic indicators demonstrated larger effect sizes, statistical significance was not achieved due to high heterogeneity or insufficient studies.

Regarding lung function, evidence suggested significant impairment among groups exposed to high-arsenic drinking water, with forced expiratory volume in one second (FEV₁) decreasing by an average of 56.97 mL (95% CI: −96.58 to −17.37) and forced vital capacity (FVC) decreasing by an average of 82.69 mL (95% CI: −140.89 to −24.49). Both indicators were classified as Class IV: weak evidence with very low GRADE ratings (n = 2,148). Overall, current evidence suggests potential associations between arsenic exposure and altered metabolic markers as well as reduced lung function, yet the strength of evidence remains limited.

Effectiveness of drinking water interventions in adults

Comprehensive evidence from existing meta-analyses indicates that interventions aimed at improving drinking-water quality effectively reduce associated health risks in adults. One high-quality meta-analysis demonstrated that various water-quality improvement measures significantly reduced diarrhea incidence (RR = 0.60, 95% CI: 0.53–0.68; Class II: highly suggestive, GRADE: very low; n = 81,215) (Figure 4). Specifically, interventions using point-of-use (POU) water treatment methods significantly lowered diarrhea risk (RR = 0.58, 95% CI: 0.51–0.66; Class II: highly suggestive, GRADE: very low; n = 72,054). Among these POU methods, filtration was the most effective (RR = 0.48, 95% CI: 0.39–0.58; n = 15,582). Other POU methods, including solar disinfection (RR = 0.62, 95% CI: 0.42–0.93; Class IV: weak evidence; n = 3,460), flocculation-disinfection treatment (RR = 0.70, 95% CI: 0.61–0.82; Class III: suggestive evidence; n = 11,788), and chlorination (RR = 0.75, 95% CI: 0.61–0.93; Class IV: weak evidence; n = 30,746), had relatively weaker effects, all three receiving very low GRADE ratings. Additionally, comprehensive water, sanitation, and hygiene (WASH) interventions among people living with HIV also significantly reduced diarrhea risk (RR = 0.57, 95% CI: 0.41–0.80; Class IV: weak evidence; n = 13,133), with a low GRADE rating.

Figure 4.

Effectiveness of drinking water interventions in adults

Summary results from re-analysis of included meta-analyses using random-effects models. Evidence Classes range from Class I (strong evidence) to Class IV (weak evidence). GRADE ratings assess evidence quality from high to very low, based on methodological limitations and consistency.

In terms of fluoride pollution control, water defluoridation interventions have highly suggestive evidence supporting their effectiveness in reducing the prevalence of dental and skeletal fluorosis. One meta-analysis revealed that defluoridation measures significantly reduced the prevalence of dental fluorosis among adults (OR = 0.26, 95% CI: 0.16–0.41, Class II: highly suggestive; n = 12,299). Likewise, skeletal fluorosis risk decreased significantly (OR = 0.33, 95% CI: 0.21–0.51; Class II: highly suggestive; n = 16,305), both with very low GRADE ratings. Furthermore, defluoridation measures significantly decreased urinary fluoride levels in adults (weighted mean difference [WMD] = −0.57, 95% CI: −0.65 to −0.49; Class III: suggestive evidence; n = 753) and fluoride concentrations in drinking water (WMD = −2.17, 95% CI: −2.91 to −1.43; Class II: highly suggestive evidence; n = 5,149), both outcomes rated very low in GRADE.

Additionally, one meta-analysis demonstrated that comprehensive water-quality improvement interventions significantly reduced fecal contamination in drinking water (OR = 0.19, 95% CI: 0.12–0.29, Class II: highly suggestive evidence, GRADE: very low; n > 1,000). Conversely, interventions based on nanoparticle water treatment technology showed only modest effectiveness in reducing antibiotic contamination in water (relative reduction = 0.58, 95% CI: 0.47–0.73; Class IV: weak evidence, GRADE: very low).

Effectiveness of drinking water interventions among children

Numerous meta-analyses have assessed the health benefits of improved drinking-water quality among children, focusing primarily on diarrhea risk and fluoride-related health outcomes (Figure 5). Overall, these interventions have shown clear protective effects. Regarding diarrhea, a meta-analysis of 83 randomized intervention studies involving 62,616 children under 5 years of age demonstrated that general water-quality improvement interventions significantly reduced diarrhea risk (RR = 0.68, 95% CI: 0.62–0.74; Class II: highly suggestive evidence), with a moderate GRADE rating. Particularly, POU water treatment interventions also showed a significant protective effect (RR = 0.58, 95% CI: 0.49–0.68, Class II: highly suggestive evidence, GRADE: low; n = 17,664). Within specific POU interventions, filtration methods showed the strongest protective effect (RR = 0.48, 95% CI: 0.36–0.64, Class II: highly suggestive evidence, GRADE: low; n = 1,634). Solar disinfection methods also effectively reduced diarrhea risk (RR = 0.68, 95% CI: 0.60–0.78, Class II highly suggestive evidence, GRADE: low; n = 1,575). Off-premises solar disinfection measures exhibited a somewhat weaker protective effect (RR = 0.70, 95% CI: 0.58–0.84, Class III: suggestive evidence, GRADE: moderate; n = 2,496). Chlorination methods significantly reduced diarrhea risk (RR = 0.62, 95% CI: 0.52–0.74, Class II: highly suggestive evidence, GRADE: very low; n = 5,131). Off-premises chlorination did not show statistically significant effects (RR = 0.82, 95% CI: 0.66–1.02; n = 6,706).

Figure 5.

Effectiveness of drinking water interventions in children

Summary results from re-analysis of included meta-analyses using random-effects models. Evidence Classes range from Class I (strong evidence) to Class IV (weak evidence). GRADE ratings assess evidence quality from high to very low, based on methodological limitations and consistency.

Water supply improvement measures showed comparatively weaker effects. Off-premises water supply improvements (n = 3,810) significantly reduced diarrhea risk (RR = 0.76, 95% CI: 0.64–0.90; Class IV: weak evidence), while on-premises water supply improvements were not significantly effective (RR = 0.83, 95% CI: 0.53–1.31). Additionally, on-premises water supply did not confer significant additional benefits compared to off-premises improvements (RR = 0.90, 95% CI: 0.67–1.21), with a very low GRADE rating.

For fluoride-related health risks, water defluoridation measures showed clear protective effects. A meta-analysis comprising 17 studies (53,419 children aged 8–15 years) demonstrated that defluoridation significantly reduced overall dental fluorosis prevalence (OR = 0.19, 95% CI: 0.11–0.35; Class II: highly suggestive evidence), although rated as very low by GRADE. Specifically, moderate dental fluorosis risk significantly decreased (OR = 0.33, 95% CI: 0.12–0.88; Class IV: weak evidence), whereas severe fluorosis showed a downward but statistically non-significant trend (OR = 0.15, 95% CI: 0.02–1.41). Moreover, defluoridation significantly increased the likelihood of children having normal tooth enamel (OR = 3.39, 95% CI: 1.29–8.93; Class IV: weak evidence). All three have very low GRADE ratings. Children’s urinary fluoride levels were significantly reduced by defluoridation interventions (SMD = −2.55, 95% CI: −3.25 to −1.84; Class II: highly suggestive evidence; 2,030 children under 5 years; WMD = −2.01, 95% CI: −3.09 to −0.94; Class III: suggestive evidence; 1,520 children aged 8–15 years), with very low GRADE ratings.

Robustness and prospective scenario analyses

We systematically evaluated 119 associations (67 related to exposures/pollution and 52 related to solutions/interventions) using robustness analyses and prospective scenario analyses (Table 1). For statistically significant associations (n = 35), Rosenthal’s fail-safe number (FSN) indicated moderate to high robustness against potential publication bias, with values ranging from 26 to 132,496. Highly robust associations included nitrate exposure and bladder cancer (FSN = 132,496), urinary total arsenic concentration (FSN = 4,559), arsenic metabolism indicators (DMA; FSN = 2,299), liver cancer mortality associated with arsenic (FSN range: 161–1,609), typhoid fever linked to microbiological contamination (FSN = 363), adverse pregnancy outcomes (FSN range: 56–198), and pulmonary function impairments (FSN range: 26–38).

Table 1.

Results of FSN and scenario analysis for robustness related to unsafe drinking water

Population	Health outcome	Evaluate the measure of water pollution (solution)	FSN	Minimum studies required (conservative scenario)	Minimum studies required (optimistic scenario)	Studies required for 80% statistical power (conservative scenario)	Studies required for 80% statistical power (optimistic scenario)
Maternity	maternal mortality	water pollution	28	–	–	–	–
General	ovary & uterine corpus cancer risk	nitrate consumption	–	8	2	18	4
General	breast cancer risk	nitrate consumption	–	9	3	20	5
General	brain & glioma risk	nitrate consumption	10	–	–	–	–
General	non-Hodgkin’s lymphoma risk	nitrate consumption	–	28	1	64	2
General	stomach cancer risk	nitrate consumption	38	–	–	–	–
General	pancreatic cancer risk	nitrate consumption	–	1	1	5	1
General	esophageal cancer risk	nitrate consumption	–	2	1	5	1
General	bladder cancer risk	nitrate consumption	–	132,496	7	Inf	9
General	kidney cancer risk	nitrate consumption	–	12	2	27	2
General	colon cancer risk	nitrate consumption	93	–	–	–	–
General	rectum cancer risk	nitrate consumption	–	12	2	30	2
General	diarrhea	EC or thermotolerant FC	–	2	1	11	1
General	diarrhea	EC	44	–	–	–	–
General	diarrhea	thermotolerant FC	–	133	2	276	3
Maternity and children	major birth defects incidence	trihalomethane exposure during pregnancy	56	–	–	–	–
Maternity and children	major birth defects incidence	arsenic exposure during pregnancy	–	8	2	17	4
General	typhoid fever	microbiologically unsafe water	363	–	–	–	–
General	CVD incidence	As	7	–	–	–	–
General	CVD mortality	As	33	–	–	–	–
General	CHD incidence	As	21	–	–	–	–
General	CHD mortality	As	60	–	–	–	–
General	stroke incidence	As	–	1	1	2	1
General	stroke mortality	As	–	21	1	47	2
General	congenital anomalies	chlorination disinfection by-products	21	–	–	–	–
General	ventricular septal defects	chlorination disinfection by-products	11	–	–	–	–
Maternity and children	spontaneous abortion	As	57	–	–	–	–
Maternity and children	stillbirth	As	198	–	–	–	–
Maternity and children	preterm delivery	As	–	6	1	13	2
Maternity and children	birth weight	As	29	–	–	–	–
Maternity and children	neonatal mortality	As	78	–	–	–	–
Maternity and children	infant mortality	As	45	–	–	–	–
General	FEV1	As	26	–	–	–	–
General	FVC	As	38	–	–	–	–
General	TAs concentration in urine	As	4,559	–	–	–	–
General	iAs concentration in urine	As	–	358	3	737	4
General	MMA concentration in urine	As	–	6	1	17	2
General	DMA concentration in urine	As	2,299	–	–	–	–
General	iAs% in urine	As	–	4,254	5	Inf	6
General	MMA% in urine	As	–	161	3	334	3
General	DMA% in urine	As	560	–	–	–	–
General	the PMI	As	293	–	–	–	–
General	the SMI	As	93	–	–	–	–
General	liver cancer mortality	iAS exposure	1,609	–	–	–	–
Male	liver cancer mortality	iAs exposure	646	–	–	–	–
Female	liver cancer mortality	iAs exposure	161	–	–	–	–
General	mortality risk of CHD	As	78	–	–	–	–
General	mortality risk of CVD	As	82	–	–	–	–
General	mortality risk of stroke	As	–	53	2	114	2
General	fatal and non-fatal risk of CHD	As	22	–	–	–	–
General	fatal and non-fatal risk of CVD	As	22	–	–	–	–
General	fatal and non-fatal risk of stroke	As	–	2	1	6	1
General	fatal and non-fatal risk of hypertension	As	18	–	–	–	–
General	fatal and non-fatal risk of carotid atherosclerosis	As	30	–	–	–	–
General	pulse blood pressure	As	–	7	2	14	3
General	QT prolongation	As	–	5	2	12	3
General	fecal contamination	all water-quality improvement	3,639	–	–	–	–
General	diarrheal morbidity	all water-quality improvement	6,107	–	–	–	–
General	diarrhea	all water-quality improvement	26,984	–	–	–	–
General	diarrhea	source water improvement	–	5	1	14	2
General	diarrhea	all POU treatment	19,808	–	–	–	–
Children <5 years	diarrhea	water quality intervention	10,053	–	–	–	–
Children <5 years	diarrhea	source water improvement	–	77	5	159	8
Children <5 years	diarrhea	POU treatment	9,896	–	–	–	–
General	diarrhea	POU: water chlorination	511	–	–	–	–
Children <5 years	diarrhea	POU: water chlorination	422	–	–	–	–
General	diarrhea	POU: flocculation and disinfection	79	–	–	–	–
General	diarrhea	POU: filtration	3,643	–	–	–	–
General	diarrhea	POU: solar disinfection	74	–	–	–	–
General	diarrhea	water supply intervention	601	–	–	–	–
People living with HIV	diarrhea	WASH interventions	417	–	–	–	–
Adults	urinary fluoride levels	water improvement	3,069	–	–	–	–
Children	diarrhea	improved, not on premises vs. unimproved	1,370	–	–	–	–
Children	diarrhea	improved, on-premises vs. unimproved	–	16	3	34	5
Children	diarrhea	improved, on premises vs. improved, not on premises	–	42	3	90	4
Children	diarrhea	improved, on premises, higher water quality vs. improved, on premises	–	2	1	6	2
Children	diarrhea	POU: chlorine	1,113	–	–	–	–
Children	diarrhea	POU: chlorine (not on premises)	–	2	1	9	1
Children	diarrhea	POU: solar disinfection	214	–	–	–	–
Children	diarrhea	POU: solar disinfection (not on premises)	68	–	–	–	–
Children	diarrhea	POU: filtration	570	–	–	–	–
Children	diarrhea	POU: filtration (not on premises)	57	–	–	–	–
Children	diarrhea	any intervention	23,474	–	–	–	–
General	water fluoride levels	water improvement and defluoridation	27,245	–	–	–	–
Children between the ages of 8 and 15 years	degree of dental fluorosis, normal	water improvement and defluoridation	233	–	–	–	–
Children between the ages of 8 and 15 years	degree of dental fluorosis, suspect	water improvement and defluoridation	–	10	2	24	2
Children between the ages of 8 and 15 years	degree of dental fluorosis, very mild	water improvement and defluoridation	–	4	1	12	1
Children between the ages of 8 and 15 years	degree of dental fluorosis, mild	water improvement and defluoridation	–	6,704	5	Inf	7
Children between the ages of 8 and 15 years	degree of dental fluorosis, moderate	water improvement and defluoridation	40	–	–	–	–
Children between the ages of 8 and 15 years	degree of dental fluorosis, severe	water improvement and defluoridation	–	2	1	6	2
Children between the ages of 8 and 15 years	urinary fluoride levels	water improvement and defluoridation	2,873	–	–	–	–
Adults	urinary fluoride levels	water improvement and defluoridation	105	–	–	–	–

FSN estimates the number of unpublished studies with zero effect sizes needed to overturn the current significant result (p < 0.05). An FSN value exceeding the threshold of 5k+10 (k represents the number of included studies) indicates that the result is robust against publication bias and not merely a chance finding based on a few positive studies. Conservative scenario, assume future studies have the same effect size as the pooled effect size of the current studies; optimistic scenario, assume future studies have the same effect size as the most influential study in the current data; minimum studies required (conservative scenario), the least number of studies needed to achieve statistical significance in the conservative scenario; minimum studies required (optimistic scenario), the least number of studies needed to achieve statistical significance in the optimistic scenario; minimum studies required represents the threshold needed for future research to ensure that the 95% confidence interval of the pooled effect size does not include the null value. Studies required for 80% statistical power (conservative scenario), the number of studies required to achieve 80% statistical power under the conservative scenario; studies required for 80% statistical power indicates a higher standard, ensuring that future studies are likely to yield a stable and reliable conclusion, with an 80% probability of detecting the assumed effect size, thus enhancing the robustness and replicability of the findings. The minimum studies required and studies required for 80% statistical power in both scenarios form the foundation of our prospective scenario analysis.

Inf, related to “infinite,” meaning exceeding the estimated range; CVD, cardiovascular disease; CHD, coronary heart disease; TAs, total arsenic; iAs, inorganic arsenic; MMA, monomethyl arsenic; DMA, dimethyl arsenic; iAs%, inorganic arsenic percentage; MMA%, monomethyl arsenic percentage; DMA%, dimethyl arsenic percentage; PMI, primary methylation index; SMI, secondary methylation index; CVD; CHD; FC, (“fecal”) coliforms.

Among the 21 non-significant exposure-related associations, conservative scenario analyses indicated that hundreds to thousands of additional studies would be required to achieve nominal significance (α = 0.05) or 80% conditional power (CP80), suggesting limited incremental benefit from further sample size increases. Conversely, optimistic scenarios indicated that many outcomes could achieve significance with only a few additional studies, highlighting potential priority areas for future research.

For intervention-related associations, statistically significant results (n = 26) generally exhibited strong robustness (FSN range: 511–27,245), notably comprehensive water-quality improvements (FSN = 26,984), interventions reducing childhood diarrhea (FSN = 23,474), POU filtration (FSN = 3,643), and defluoridation measures (FSN range: 105–27,245). Among the 10 non-significant intervention associations, optimistic scenarios suggested that nominal significance could often be reached with only a few additional studies (1–3 studies), such as interventions improving at-home water quality and certain dental fluorosis subtypes. However, some interventions, including POU water-source improvements targeting children and mild dental fluorosis outcomes, would require substantially more studies under conservative assumptions, indicating diminishing returns if effect sizes remain unchanged.

Discussion

This study systematically synthesized the global evidence on health outcomes and interventions related to unsafe drinking water, addressing a critical gap in the existing literature. By re-examining 25 systematic reviews and meta-analyses encompassing 158 unique outcomes, we identified significant associations between unsafe drinking water and various adverse health consequences. Concurrently, interventions to improve drinking-water quality were generally effective in mitigating these health risks. However, the evidence is inconsistent. Many contaminant effects on health are statistically significant but have low or very low GRADE certainty, whereas intervention effects, especially on childhood diarrhea, have higher certainty. This discrepancy suggests that public health strategies should prioritize scaling safe-water interventions supported by higher-certainty evidence, while underscoring the need for high-quality research to better establish causal effects of specific contaminants on health.

Our findings indicate that microbial contamination of drinking water is associated with an increased risk of enteric infectious diseases, which align well with established knowledge that contaminated water sources commonly transmit enteric diseases such as dysentery and typhoid. However, findings related to diarrhea risk show notable inconsistencies. When the exposure definition included thermotolerant coliforms, pooled analyses did not show statistically significant associations.¹⁰ This variability may stem from differences in contamination indicators, studied populations, and geographical contexts, suggesting that reliance solely on coliform counts to assess diarrhea risk is uncertain. Furthermore, although statistical associations with increased diarrhea and typhoid risks were identified, these findings predominantly derive from observational studies, which are vulnerable to bias and confounding factors, resulting in very low GRADE ratings.

Regarding chemical contamination, agents such as arsenic, fluoride, and nitrate are primary contributors to chronic diseases from unsafe drinking water. For arsenic, it inflicts multi-organ damage through pathways including oxidative stress, abnormal DNA methylation, and endothelial injury. Moreover, the International Agency for Research on Cancer (IARC) has classified arsenic in drinking water as a Group 1 carcinogen.¹¹ Our research reveals that high-level arsenic exposure significantly increases the risk of liver cancer mortality by 80%, with a more pronounced effect in males.¹² It also promotes atherosclerosis and is associated with a trend of increased risk of coronary heart disease and hypertension.¹³ Furthermore, prenatal exposure leads to arsenic accumulation in the placenta, elevating the risks of miscarriage, stillbirth, and perinatal mortality.⁴ However, due to the predominance of case-control designs and prevalent bias in the original studies, which inherently present significant challenges in controlling for confounding factors such as hypertension, dyslipidemia, and smoking, our GRADE assessment downgraded the quality of this evidence to “very low.”

Of particular note, our analysis identified a significant increase in cardiovascular disease risk even at low arsenic exposure levels approaching the current World Health Organization (WHO) guideline (10 μg/L). Compared to a baseline of 1 μg/L, the relative risk was 1.41, consistent with previous meta-analysis findings by Xu et al.¹³ This indicates that current standards may not fully eliminate cardiovascular risks from low-level arsenic exposure, posing ongoing challenges to global drinking-water regulation. These results underscore the importance of further research into the mechanisms linking arsenic exposure with cardiovascular and metabolic conditions. Additionally, they highlight that health hazards associated with unsafe drinking water extend beyond traditionally recognized outcomes, requiring a more comprehensive and integrated assessment approach.

The health impacts of high-fluoride drinking water were also evident in our study. We found that long-term consumption of high-fluoride water by children was associated with impaired cognitive development; the average IQ in the exposed group was significantly lower than in the non-exposed group.^14,15 Statistically, we classified this as Class II: highly suggestive evidence, a finding that aligns with previous research indicating the potential adverse effects of fluoride on childhood cognitive development.⁷ However, because the primary research in this field is geographically concentrated in high-fluoride regions and predominantly employs cross-sectional survey designs, leading to significant heterogeneity and potential for confounding, the GRADE rating for this evidence was downgraded to “very low.” This indicates that a causal relationship cannot be confidently established at present. Regardless, the public health significance of this finding is clear: while water fluoridation has traditionally focused on the benefits for dental health, we must also be mindful of the potential neurodevelopmental risks associated with excessive fluoride intake. Policymaking should, therefore, involve a careful weighing of these benefits and risks and be supported by more high-quality research to confirm these findings. In parallel, established studies indicate that chronic consumption of water with excessive fluoride levels disrupts the normal mineralization processes of bones and teeth, which explains our observation of an increased risk of skeletal and dental fluorosis in both adults and children in areas with high fluoride exposure.

Nitrate contamination, on the other hand, primarily increases cancer risk through its in vivo conversion to nitrite, which can then form carcinogenic N-nitroso compounds.¹⁶ Our study observed that elevated concentrations of nitrate in drinking water were associated with a slight increase in the risk of colon cancer, stomach cancer, and brain tumors. However, the effect sizes were small, the evidence exhibited high heterogeneity, and the statistical strength was weak, resulting in a GRADE rating of “very low.”¹⁷ This is likely because the carcinogenic effect of nitrate is relatively weak and is easily confounded by other factors, such as diet, which are difficult to control for in existing studies. This conclusion is consistent with the assessment by the IARC, which classifies nitrate/nitrite that is ingested and forms N-nitroso compounds in vivo as Group 2A carcinogens, acknowledging that there is limited evidence in humans but sufficient evidence in experimental animals.

Another finding of our study is that high-level exposure to drinking-water disinfection by-products (DBPs), such as THMs, is associated with an increased risk of bladder and rectal cancer, corroborating the known carcinogenic potential of chlorine-based DBPs.¹⁸ Furthermore, our analysis identified several associations supported by limited but suggestive evidence, warranting further investigation. For example, high concentrations of chlorinated DBPs may be linked to birth defects and contaminated drinking water might increase maternal mortality risk, especially in regions lacking adequate medical infrastructure.^19,20 This indicates that water contamination could elevate mortality by increasing infections or pregnancy-related complications. Although these associations originate primarily from observational studies, they consistently appear across diverse populations and outcomes and are biologically plausible. Therefore, ensuring microbial safety while simultaneously managing DBP concentrations involves carefully balancing associated risks and benefits.

Our research also suggests that improving drinking-water quality effectively reduces various health risks. A high-quality meta-analysis demonstrated that water-quality improvement interventions collectively reduce diarrhea risk by approximately 40%, classified as Class II (highly suggestive evidence) despite a “very low” GRADE rating, largely due to methodological limitations of primary studies.²¹ Specifically, POU treatment measures reduced diarrhea risk by about 42%, with filtration-based methods showing the strongest effect (52% reduction). Other methods such as solar disinfection (SODIS), flocculation disinfection, and chlorination also provided some protective effects, although classified as weaker evidence (Class III–IV).²¹ Notably, comprehensive interventions combining safe water, sanitation, and hygiene significantly reduced diarrhea among immunocompromised populations, including individuals with HIV. Despite bias and heterogeneity in available studies, these findings clearly highlight the benefits of drinking water quality interventions in preventing infectious diseases like diarrhea.

In the prevention and control of fluoride contamination, water defluoridation interventions produce substantial health benefits. Multiple meta-analyses demonstrate that improving water sources and implementing defluoridation effectively reduce dental and skeletal fluorosis risks in adults (Class II: highly suggestive evidence), significantly lowering overall fluoride burden.²² Despite the evidence’s “very low” GRADE rating, these findings clearly indicate defluoridation’s critical role in fluorosis prevention. Improving drinking-water quality also significantly reduces microbial contamination risks. However, evidence remains limited for treatments targeting emerging contaminants. For example, preliminary studies indicate only moderate efficacy of nanomaterial-based water treatments for removing antibiotics, with weak evidence strength (Class IV, very low GRADE).²³ Thus, while traditional methods to control microbial contamination are well supported, treatment technologies for emerging contaminants require further high-quality validation research.

For children, improvements in drinking-water quality also show clear health-protective effects. A large-scale meta-analysis confirmed that various drinking-water interventions collectively reduce the risk of diarrhea in preschool-aged children by approximately 32%, with a highly suggestive evidence level (Class II) and a moderate GRADE rating.²⁴ Among these, POU water treatment is particularly effective for children, reducing the average risk of diarrhea by about 42%. Among the specific POU measures, household filtration can decrease the incidence of childhood diarrhea by around 50%, while SODIS and household chlorination can lower the risk by approximately 30%–37%.²⁴ In contrast, merely improving the water supply source, while offering some risk reduction, did not confer an additional advantage, suggesting that simply increasing the convenience of water access is insufficient to further reduce the burden of childhood diarrhea.

Water defluoridation measures significantly reduce health risks associated with fluoride exposure in children. Comprehensive analyses of children and adolescents aged 8–15 years showed that long-term defluoridation effectively lowers the overall prevalence of dental fluorosis (Class II, highly suggestive evidence) and significantly decreases children’s urinary fluoride levels by approximately 2 mg/L compared to controls. Despite the “low” or “very low” GRADE rating due to risk of bias, these findings support defluoridation as an effective intervention for preventing dental fluorosis and reducing fluoride exposure.²² Notably, our analysis also indicates that improvements in water supply infrastructure alone may not consistently yield expected health benefits. While centralized community water sources (e.g., water stations, wells) can moderately reduce diarrhea risk, providing piped water directly to households did not show significantly greater benefits compared to community-based water points.²⁴ Thus, current evidence suggests that simply increasing convenience through household-level piped water supply is insufficient for further significant reductions in diseases such as diarrhea.

Crucially, a core finding of this umbrella review revealed a key feature of the evidence base in the field of drinking-water safety: a significant discrepancy between the uncertainty of risk attribution and the certainty of intervention effectiveness. In terms of health risks, this study identified several statistically highly significant contaminant-health associations. However, when assessed using the GRADE framework, the overall certainty of this evidence was overwhelmingly rated as “low” or “very low.” This phenomenon is not a limitation of our study but an inherent feature revealed by synthesizing the current evidence landscape. It stems from the fact that evidence classification frameworks primarily focus on statistical metrics such as sample size, p values, and heterogeneity. In contrast, the GRADE framework provides a more comprehensive assessment of the credibility of evidence, incorporating factors like risk of bias, inconsistency, indirectness, and imprecision. The evidence for contaminant exposure-outcome associations is predominantly derived from observational studies. Such study designs face inherent and substantial challenges in controlling for confounding factors, accurately quantifying long-term, low-dose exposures, and avoiding selection bias. Consequently, even when certain associations are statistically significant, this is insufficient to elevate the certainty of the evidence; our confidence in the causal nature and precision of these effect estimates remains low due to the methodological weaknesses of the primary research. For example, this review found a “convincing” (Class I) association between arsenic exposure in drinking water and cardiovascular disease, yet the GRADE rating was still “very low.” This is because, while the trend was consistent across populations, the supporting evidence consisted mostly of observational analyses and there were signs of bias in the included studies. Furthermore, although the association between high level of fluoride in drinking water and lower childhood IQ showed a large effect size, we could only classify it as “suggestive evidence” with a “very low” GRADE rating due to high heterogeneity among studies and ongoing debate regarding the direction of causality.

In contrast to the uncertainty surrounding health risk evidence, evidence supporting intervention effectiveness is substantially more robust. We identified 13 instances of highly suggestive (Class II) evidence for effective interventions, with three outcomes reaching a “moderate” GRADE certainty. In particular, improving drinking-water quality significantly reduces childhood diarrhea risk. POU water treatments, especially filtration, consistently demonstrate effectiveness across all age groups. Defluoridation interventions similarly show highly suggestive evidence for reducing dental and skeletal fluorosis, supported by consistent protective effects from multiple independent trials with large sample sizes and relatively lower bias risks. Under optimal conditions, POU filtration can reduce childhood diarrhea risk by as much as 52%. However, translating this efficacy into large-scale, sustainable, real-world “effectiveness” faces a significant implementation gap. Even low-cost interventions, such as chlorination products or simple filters, can represent a considerable expense for extremely impoverished households, potentially causing such interventions to exacerbate rather than narrow health inequalities. Behavioral and cultural factors also influence the long-term, correct use of interventions and the adoption of new technologies. Moreover, project support and sustainability are major challenges to scaling up. The effectiveness of drinking-water interventions is also influenced by the broader WASH context. In environments with extremely poor sanitation or severely contaminated water sources, a standalone water quality intervention may have limited impact. Consequently, while many interventions in this study show large effect sizes, their GRADE ratings are still “low” or “very low.” This largely reflects the substantial heterogeneity across different studies, which arises from the implementation factors mentioned above. This indicates that the focus of policy should not merely be on promoting a specific technology but must be on building a socio-technical system that can support the sustained and correct use of that technology.

In summary, when discussing the results, this study rigorously distinguishes between the strength of a statistical association and the reliability or certainty of the evidence. This evidence dichotomy allows for the use of large-scale, aggregated data to identify risk signals while simultaneously avoiding over-interpretation based on low-quality evidence. This methodological approach also leads to the phenomenon where, despite statistically significant associations, our confidence in a causal relationship (e.g., the GRADE certainty) remains low. This is not to deny the existence of these risks; rather, it forms the central narrative for understanding the current challenges in global drinking-water safety. It suggests that public health decision-making should not be based solely on the precise quantification of specific contaminant risks but should instead be guided by the certain benefits offered by known, effective solutions.

However, as an umbrella review, our conclusions are inherently constrained by the methodological quality of the included meta-analyses and their underlying primary studies. Many original studies lacked pre-registration and adequate bias control, resulting in predominantly “low” or “critically low” AMSTAR 2 ratings. This underscores an urgent need for more standardized, high-quality meta-analyses. Furthermore, substantial statistical heterogeneity (I² > 50%) observed in many pooled estimates likely reflects differences in study populations, exposure levels, intervention fidelity, and study designs, necessitating cautious interpretation despite our use of random-effects models. Publication bias remains a concern, as studies with smaller or non-significant effects may remain unpublished, potentially inflating observed associations. Additionally, our review assessed the health effects of single pollutants only, unable to examine the real-world complexities of combined chemical exposures due to limited primary data. Crucially, the contrast identified by our study, uncertain risk attribution versus relatively certain intervention benefits, offers clear guidance for policymakers. Waiting for definitive causal evidence linking every contaminant to specific health outcomes before taking action is neither practical nor necessary. From public health and economic perspectives, investing in proven interventions that prevent disease outweighs prolonged uncertainty over exact risk quantification. Hence, policy should prioritize solution implementation over continued risk debate. Finally, the geographical scope is limited by existing published meta-analyses, potentially underrepresenting health risks and intervention outcomes in specific developing regions or marginalized communities.

Unsafe drinking water is not merely an environmental problem; it is a profound social issue and a significant manifestation of health inequity.²⁵ It acts as a social determinant of health that disproportionately affects poor, rural, marginalized, and minority communities. These populations often rely on untreated groundwater and lack both the economic capacity and the political voice to secure access to safe water sources, all while shouldering a heavier burden of disease. Therefore, water security policy must be centered on health equity. This means that resource allocation should be skewed toward the most vulnerable populations, and the design of interventions must consider their specific needs and ability to pay. The objective must be to close the health gap, not simply to raise the average standard. Furthermore, the issue of water security should not be addressed in isolation but must be integrated into broader public health and development agendas. Water-quality improvement interventions should be combined with programs focused on child health, nutrition improvement, and other areas to achieve synergistic effects.²⁶ For example, supplementing child nutrition programs in regions with a high incidence of diarrhea with a safe drinking-water supply can simultaneously improve nutrient absorption and prevent disease. Policymakers need to recognize that there is no “one-size-fits-all” solution. An effective national strategy should comprise a “policy toolbox” with a variety of options. This approach would empower local authorities to select the most suitable interventions from a range of evidence-based measures, from behavioral change communication and subsidies for POU products to large-scale infrastructure investment and strict regulatory enforcement, based on their own risk assessments, economic conditions, and cultural contexts.

To effectively translate the findings of this umbrella review into policy recommendations, we applied a simplified EtD framework. Policy priorities are established by assessing disease burden, evidence strength, effect sizes, and cost-effectiveness, with additional emphasis placed on interventions targeting vulnerable populations or those providing substantial benefits at relatively low cost. Policymakers are advised to prioritize high-quality evidence generation in areas currently lacking robust data, incorporate sociological considerations (equity, acceptability, and feasibility), and establish continuous monitoring and feedback mechanisms to regularly refine policy strategies. Developing a sustainable information platform that aggregates updated evidence and implementation costs would further support iterative decision-making and encourage cross-sectoral collaboration. Table 2 provides illustrative EtD analyses of key findings, demonstrating how various factors can guide specific policy recommendations.

Table 2.

Simplified EtD framework for key findings on the risks and interventions related to unsafe drinking water

Health outcomes and population at risk	Pollutants and effect sizea	Certainty of evidenceb	Recommended intervention planc	Cost estimate/cost-effectivenessd	Policy priority and basis for determinatione
Cardiovascular disease
Incidence and mortality risk of cardiovascular disease (general population)	arsenic exposure >10 μg/L vs. <1 μg/L; RR, 1.41 (95% CI: 1.25–1.60)	evidence class: Class I: convincing evidence	1. conduct widespread screening of arsenic concentrations in water sources and identify safe wells.	costs of arsenic removal vary greatly: government-led deep well projects are expensive (∼US$142 per person)26; however, low-cost informational interventions (water quality testing and switching well guidance) cost less than US$10 per person.27	Monitoring and targeted interventions. Rationale: although the GRADE rating is very low, Class I statistical evidence strongly indicates a real risk. Given the enormous disease burden of cardiovascular disease, any clearly established risk factor has major public health implications. Low-cost, highly effective monitoring and informational interventions should be prioritized, with targeted technical interventions in high-risk areas.
		AMSTAR2: very low	2. in highly polluted areas, promote community-level arsenic removal facilities (e.g., activated alumina, ferric salt co-precipitation) or switch to deeper groundwater.
		GRADE: very low	–
Cancer
Risk of death from liver cancer (general population)	inorganic arsenic exposure (yes vs. no); SMR, 1.80 (95% CI: 1.53–2.10)	evidence class: Class II: highly suggestive evidence	same as above, with emphasis on water-source replacement or community-level arsenic removal in high-arsenic regions.	same as above	Monitoring and targeted interventions. Rationale: Class II evidence and a large effect size (80% increased risk) provide a strong risk signal. Because liver cancer is highly fatal, the value of prevention is immense. Given the low GRADE rating, broad, high-cost interventions should be approached cautiously, but monitoring and interventions in known high-risk areas are justified.
		AMSTAR2: very low
		GRADE: very low
Incidence risk of colon cancer (general population)	nitrate exposure (high vs. low); OR, 1.15 (95% CI: 1.04–1.28)	evidence class: Class IV: weak evidence	1. strengthen monitoring of nitrate levels in drinking water in agricultural areas 2. promote water source protection policies and control the use of nitrogen fertilizers. 3. consider ion exchange or reverse osmosis treatment in areas with particularly high concentrations.	there is a lack of standardized cost-effectiveness data.	Research and monitoring should be prioritized. Rationale: both the evidence class and GRADE rating are very low, and the effect size is small. Although the burden of colorectal cancer is heavy, the current evidence is insufficient to support large-scale, high-cost interventions. Therefore, the core tasks are to strengthen monitoring to identify hotspot areas and to conduct high-quality epidemiological studies to clarify causal relationships.
		AMSTAR2: low
		GRADE: very low
Incidence risk of brain glioma (general population)	nitrate exposure (high vs. low); OR, 1.16 (95% CI: 1.08–1.26)	evidence class: Class III: suggestive evidence	same as above, focusing primarily on monitoring and source control.	same as above	Research and monitoring should be prioritized. Rationale: similar to colorectal cancer, although the statistical evidence is slightly stronger (Class III), the GRADE rating is very low and the associated biological mechanisms are not fully understood. At present, efforts should focus on research and monitoring.
		AMSTAR2: low
		GRADE: very low
Maternal and child health
Impaired cognitive function in children (IQ scores)	fluoride exposure (children with fluorosis vs. normal children); SMD, −10.98 (95% CI: −13.82 to −8.14)	evidence class: Class II: highly suggestive evidence	1. mandate defluoridation of drinking water in high-fluoride areas.	costs of community-level defluoridation vary by technology and scale, with a wide range of estimates. Low-cost adsorbent materials (e.g., bauxite and red clay) and ECF show potential in developing countries.28	Targeted interventions and prioritized research. Rationale: the effect size is enormous (a decrease of nearly 11 IQ points), which has major implications for individual and societal development. Although the GRADE rating is very low, interventions in known high-fluoride areas are justified on the basis of the precautionary principle and the severity of potential harm. At the same time, high-quality research is urgently needed to increase the certainty of the evidence.
		AMSTAR2: very low	2. promote community-level or household defluoridation equipment (e.g., activated alumina, bone char).
		GRADE: very low	–
Neonatal mortality (pregnant population)	maternal arsenic exposure during pregnancy (high vs. low); OR, 1.51 (95% CI: 1.26–1.82)	evidence class: Class III: suggestive evidence	provide pregnant women with reliable access to safe drinking water, especially in known arsenic-contaminated areas. Low-cost household water purification devices can be promoted.	same as arsenic contamination interventions. Targeted small-scale interventions for pregnant women may be more cost-effective.	Targeted interventions and prioritized research. Rationale: the evidence class is III and the GRADE rating is “low,” the effect size is significant (51% increased risk), and the target population (newborns) is extremely vulnerable. This is a clear intervention signal; pregnant women in high-risk areas should be given priority access to safe drinking water, and high-quality research is urgently needed to increase the certainty of the evidence.
		AMSTAR2: very low
		GRADE: very low
Stillbirth (pregnant population)	maternal arsenic exposure during pregnancy (high vs. low); OR, 1.75 (95% CI: 1.33–2.31)	evidence class: Class III: suggestive evidence	same as above	same as above	Targeted interventions. Rationale: similar to neonatal mortality risk, the effect size is large (75% increased risk) and the evidence class is III, both pointing to the serious harm of arsenic exposure during pregnancy. The same intervention strategies as those for neonatal mortality risk should be adopted.
		AMSTAR2: very low
		GRADE: very low
Infectious diseases
Incidence risk of typhoid (general population)	exposure to microbiologically unsafe water; OR, 2.78 (95% CI: 1.80–4.32)	evidence class: Class III: suggestive evidence	improve water-source sanitation, promote safe drinking habits and vaccination. POU treatment (filtration, disinfection) can effectively remove pathogens.	the cost-effectiveness can be referenced from interventions for diarrhea.	Act in concert with diarrhea interventions. Rationale: the effect size is enormous (nearly double the risk), and the etiology and interventions overlap greatly with diarrheal disease. It should be incorporated into broader water sanitation improvement and POU promotion programs.
		AMSTAR2: very low
		GRADE: very low
Diarrheal disease
Incidence rate of diarrhea in children (<5 years)	POU household filtration; RR, 0.48 (95% CI: 0.36–0.64)	evidence class: Class II: highly suggestive evidence	vigorously promote and subsidize efficient POU filters (such as ceramic and hollow-fiber membrane filters), along with maintenance and hygiene education.	due to differences in technologies and regions, cost-effectiveness ranges from a few dollars to several hundred dollars per DALY, but is generally highly cost-effective.29	Universal promotion. Rationale: high disease burden, a large effect size (52% risk reduction), moderate-quality evidence, and extremely high cost-effectiveness. This is one of the most well-established, highest return interventions at present, and should be among the top priorities for global child health and water safety policies.
		AMSTAR2: very low
		GRADE: moderate
Incidence rate of diarrhea in children (<5 years)	POU household chlorination; RR, 0.62 (95% CI: 0.52–0.74)	evidence class: Class II: highly suggestive evidence	promote community-level automatic chlorination devices (dispensers) or provide low-cost household bleach, accompanied by social marketing on proper use.	specific cost-effectiveness data are lacking, but the cost of chlorination is extremely low and overall it is highly cost-effective.30	Context-specific promotion. Rationale: similar to POU filtration, but chlorination results in an unpleasant odor and lower acceptability. It can serve as a first-choice option in specific contexts and should be promoted selectively.
		AMSTAR2: very low
		GRADE: low
Incidence rate of diarrhea in children (<5 years)	POU SODIS; RR, 0.68 (95% CI: 0.58–0.79)	evidence class: Class II: highly suggestive evidence	in areas with abundant sunlight and lack of other resources, promote this as a zero-cost alternative, with enhanced training on proper practices (e.g., duration of sun exposure, container requirements).	material costs are almost zero; the main costs are social mobilization and training.	Context-specific promotion. Rationale: similar to POU filtration and almost no economic cost. However, its effectiveness is highly influenced by climatic conditions and user compliance. It is suitable as a basic intervention in specific impoverished, remote, sun-rich areas.
		AMSTAR2: very low
		GRADE: moderate
Fluorosis
Prevalence of permanent dental fluorosis in children (8–15 years)	water-source improvement and defluoridation; OR, 0.19 (95% CI: 0.11–0.35)	evidence class: Class II: highly suggestive evidence	in known high-fluoride areas, implement community-level centralized defluoridation or switch to low-fluoride water sources.	specific cost data are lacking, but it can prevent more severe skeletal fluorosis and potential neurotoxicity, yielding significant long-term benefits.	Targeted interventions and prioritized research. Rationale: the intervention effect is very strong (81% risk reduction) and the evidence class is II. Although the GRADE rating is low and dental fluorosis itself is a relatively minor burden, it is a marker for more serious health risks (skeletal fluorosis, cognitive impairment). Therefore, defluoridation interventions in high-fluoride areas are a key upstream measure to prevent a range of health problems, and high-quality research is urgently needed to enhance the certainty of the evidence.
		AMSTAR2: high
		GRADE: very low
Urinary fluoride levels in children	water-source improvement and defluoridation; SMD, −2.55 (95% CI: −3.25 to −1.84)	evidence class: Class II: highly suggestive evidence	same as above	same as above	Targeted interventions and prioritized research. Rationale: the large effect size and Class II evidence indicate that defluoridation interventions can very effectively reduce human fluoride burden. This provides strong evidence supporting the aforementioned intervention recommendations for dental fluorosis and cognitive impairment, but high-quality research is still needed to increase the certainty of the evidence.
		AMSTAR2: very low
		GRADE: very low

ECF, electrocoagulation-flotation; DALY, disability-adjusted life year.

^aPollutants/interventions and effect sizes: data are derived from the re-analysis conducted in this umbrella review.

^bCertainty of the evidence: based on a combination of the statistical evidence grading (Class I–IV) and the GRADE quality rating.

^cRecommended interventions: derived from effective strategies identified in this study and current public health practice recommendations.

^dEstimated cost-effectiveness: informed by existing published evaluations, providing an approximate reference range.

^ePolicy priority and decision basis: determined by integrating evidence strength, disease burden, intervention effects, and cost-effectiveness.

In conclusion, this umbrella review has comprehensively revealed the multiple threats posed by unsafe drinking water to human health and has validated the critical role of water improvement measures in disease prevention. Although the quality of much of the current evidence for specific associations is not yet ideal, we can reasonably conclude that ensuring drinking-water safety is essential for reducing the burden of infectious diseases, preventing certain chronic illnesses, and protecting the health of vulnerable populations. Policymakers should act on this evidence to accelerate the development of safe water supply infrastructure and the implementation of community-level interventions on a global scale, directing limited resources toward the most evidence-based and cost-effective solutions. Concurrently, the scientific community must strive to enhance the quality and scope of relevant research. This includes conducting more prospective and retrospective studies to investigate and clarify the harms of contaminants to the human body and their dose-response relationships. We also need more standardized, high-quality meta-analyses to improve the reliability of the evidence. Only by ensuring a strong interface between scientific evidence and public policy can the world move more rapidly toward the goal of “safe drinking water for all.” This will not only reduce preventable diseases and deaths but also will promote social equity and sustainable development, bringing well-being to billions of people.

Limitations of the study

This umbrella review has several inherent methodological limitations, primarily stemming from the nature of environmental health evidence, which is predominantly based on observational studies. To systematically evaluate this evidence, we utilized multiple approaches, including the GRADE framework. However, it is important to acknowledge that the GRADE system was originally designed within a clinical intervention context and thus assigns observational studies an initial “low” certainty rating. In environmental health research, this inherently creates a methodological ceiling, making it challenging even for well-conducted observational studies to achieve higher certainty ratings comparable to randomized controlled trials. Consequently, this limitation can restrict the accurate identification of genuine health risks.

Considering this methodological issue, we supplemented our analysis with additional robustness assessments, specifically the FSN and prospective scenario analyses, to further examine the statistical stability of findings rated as “low” or “very low” certainty by GRADE. These analyses indicated that several associations rated as having low certainty, nonetheless, exhibited substantial statistical robustness, meaning that a large number of hypothetical negative studies would be required to alter these conclusions. This suggests the necessity of distinguishing between the evidence “certainty” assigned by rating frameworks and the statistical robustness of associations when interpreting these results.

Therefore, our review provides a complementary evaluation of evidence from multiple analytical perspectives, aiming to offer a more comprehensive understanding of the characteristics revealed by different evaluation frameworks. This multidimensional approach may help policymakers and implementation researchers better interpret findings, particularly where traditional rating systems exhibit methodological constraints. Nevertheless, caution remains necessary when interpreting these conclusions, considering residual heterogeneity among studies, the complexities associated with real-world combined chemical exposures, and the limitations in geographical coverage and representativeness of available research.

Resource availability

Lead contact

Further information and requests for resources should be directed to the lead contact, Xuelei Ma (drmaxuelei@gmail.com).

Materials availability

This study did not generate new materials.

Data and code availability

• •All data from this study are presented in the supplemental information. This paper analyzes existing, publicly available data. Access to the data is available from the public literature.
• •This study did not generate any unique code.
• •Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request.

Acknowledgments

This research did not receive any specific grants from funding agencies in the public, commercial, or not-for-profit sectors.

Author contributions

S.L., X.M., and X.W. designed the study. S.L., Y.W., J.Z., and Y.L. conducted the literature search, data extraction, and quality assessment. S.L., Y.W., and J.Z. performed statistical analysis and visualization. L.Q., D.Y., Y.J., and Z.J. assisted in data verification. S.L., Y.W., and J.Z. wrote the original draft. S.L., Y.L., X.M., and X.W. supervised the study and revised the manuscript. All authors had full access to all data and accepted responsibility for submission.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
The results of health study	Benova et al.27	Data S1
The results of health study	Essien et al.17	Data S1
The results of health study	Gruber et al.10	Data S1
The results of health study	Issanov et al.28	Data S1
The results of health study	Jauniaux et al.19	Data S1
The results of health study	Mogasale et al.29	Data S1
The results of health study	Moon et al.3	Data S1
The results of health study	Nieuwenhuijsen et al.20	Data S1
The results of health study	Quansah et al.4	Data S1
The results of health study	Sanchez et al.30	Data S1
The results of health study	Shen et al.31	Data S1
The results of health study	Wang et al.15	Data S2
The results of health study	Wang et al.12	Data S2
The results of health study	Xu et al.13	Data S2
The results of intervention study	Arnold et al.32	Data S2
The results of intervention study	Bain et al.33	Data S2
The results of intervention study	Cairncross et al.34	Data S2
The results of intervention study	Clasen et al.35	Data S2
The results of intervention study	Clasen et al.21	Data S2
The results of intervention study	Fewtrell et al.36	Data S2
The results of intervention study	Malakootian et al.23	Data S2
The results of intervention study	Peletz et al.37	Data S2
The results of intervention study	Wang et al.15	Data S2
The results of intervention study	Wolf et al.24	Data S2
The results of intervention study	Wang et al.22	Data S2
Software and algorithms
R (version 4.3.2)	https://www.r-project.org/	R (version 4.3.2)

Experimental model and study participant details

We included 25 systematic reviews and meta-analyses, covering 158 unique outcome measures, all of which were re-analyzed. Detailed characteristics of the included studies are provided in Datas S1 and S2.

Method details

This study employed an umbrella review approach aiming to systematically and comprehensively integrate existing meta-analytic evidence to clarify the credibility of evidence regarding the associations between unsafe drinking water contaminants and health risks, as well as the effectiveness of interventions. The study rigorously adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Data S5) and PRISMA Abstract standards (Table S1) and was registered on PROSPERO, the International Prospective Register of Systematic Reviews (registration number: CRD420251001778), on 1 March 2025, ensuring transparency and reproducibility of the research process.³⁸

Search strategy and inclusion criteria

This study systematically searched PubMed, EMBASE, Web of Science, and the Cochrane Library databases from their inception up to 15 July 2025, with no language restrictions applied during the search. The search strategy included Medical Subject Headings (MeSH) terms and free-text keywords; full search strings, retrieval dates, and detailed search strategies for each database are provided in the supplementary materials. Search results underwent rigorous deduplication before independent screening and data extraction by two reviewers (Li S and Wei Y). Extracted data included authors, publication year, study design, sample size, contaminant type, effect sizes, and health outcome measures. Any discrepancies were resolved through arbitration by a third reviewer (Wei X).

Inclusion criteria were: systematic reviews with completed meta-analytic processes; no restrictions on demographic characteristics such as age, gender, or geographic location; exposure groups defined as populations exposed to contaminated drinking water; observational studies or randomized controlled trials (RCTs) explicitly reporting health outcomes; and studies assessing relationships between unsafe drinking water exposure or water-treatment interventions and health risks, clearly reporting associated effect sizes and statistical measures. Meta-analyses with significant methodological flaws or lacking original data support were excluded. Duplicate meta-analyses were defined as those assessing the same drinking water contaminant and health outcome. When two or more duplicate meta-analyses were identified, the one with the highest methodological quality based on AMSTAR 2 ratings was selected first; if quality ratings were equivalent, the most recent review with the largest sample size was chosen. None of the excluded duplicate meta-analyses were presented in the main text results; however, their effect sizes were recalculated and fully detailed results were provided in the appendices for reader reference. Additionally, potential duplicate weighting was addressed during data re-synthesis by performing deduplication at the individual study segment level.

Methodological quality assessment

We employed the revised AMSTAR 2 tool to assess the methodological quality of the included meta-analyses. The AMSTAR 2 tool comprises 16 evaluation items, seven of which are categorized as critical items.³⁹ Each item was rated as “Yes” (fully compliant), “Partial Yes” (partially compliant), or “No” (non-compliant). Based on compliance with critical items and the degree of fulfillment of non-critical items, study quality was classified into four levels: high, moderate, low, and critically low. After careful deliberation by our research team, item 7 of the AMSTAR 2 tool, “whether a list of excluded studies was provided”, was modified from a critical item to a non-critical item. The rationale for this adjustment is that both the PRISMA guidelines and practical meta-analysis literature often omit lists of excluded studies.⁴⁰ Thus, treating this item as non-critical better aligns with current methodological evaluation practices for meta-analyses. Retaining this item as critical might inappropriately downgrade the methodological quality of the included reviews, negatively impacting the overall quality assessment of results. Further details are provided in the supplemental information.

Evidence grading criteria

Regarding evidence grading, this study utilized the widely adopted Evidence Classification framework, categorizing credibility of findings into five levels: convincing (Class I), highly suggestive (Class II), suggestive (Class III), weak (Class IV), and non-significant (NS).^41,42 The evaluation dimensions included study sample size, significance level of effect sizes under a random-effects model, heterogeneity, publication bias, excess significance bias, and robustness of the effect size from the largest included study (for detailed scoring criteria, see the appendix).

Simultaneously, the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) framework was employed for a comprehensive evaluation of evidence quality, incorporating dimensions such as risk of bias, consistency of data, and directness.⁴³ In accordance with the GRADE methodology, we systematically assessed evidence quality based on five established domains: (i) risk of bias, encompassing limitations in study design, inadequate allocation concealment, insufficient blinding, and selective outcome reporting; (ii) inconsistency, reflected by unexplained heterogeneity or substantially divergent effect estimates; (iii) indirectness, arising from variations in populations, interventions or exposures, comparators, or outcomes relative to our predefined research questions; (iv) imprecision, characterized by small sample sizes or wide confidence intervals that cross decision-relevant thresholds; and (v) publication bias, indicated by evidence of small-study effects or funnel plot asymmetry suggesting selective dissemination of findings. Additionally, observational evidence was considered for potential upgrading based on three criteria: substantial magnitude of effect, presence of a dose-response relationship, and the likelihood that plausible residual confounding would attenuate rather than exaggerate the observed associations. Evidence quality was graded into four levels: high, moderate, low, and very low. Two reviewers (Li S and Wei Y) independently conducted the GRADE assessments, with disagreements resolved through arbitration by a third reviewer (Ma X). Detailed GRADE assessment results are available in the supplemental information.

It is important to emphasize that the Evidence Classification framework primarily focuses on quantitative measures such as statistical significance, effect size, and robustness, whereas the GRADE framework emphasizes broader dimensions of overall evidence quality, including risk of bias, heterogeneity, and directness of evidence. Consequently, evidence rated highly in statistical terms (such as Class I or II) might still be downgraded to low or very low quality under the GRADE criteria due to issues like risk of bias or heterogeneity. By explicitly highlighting the differing meanings and potential conflicts between these two grading systems, this study aims to remind policymakers and researchers to interpret and apply findings cautiously.

Evidence-to-decision framework

Finally, to effectively translate evidence into policy recommendations, this study incorporated the EtD framework recommended by the WHO and the GRADE Working Group.⁴⁴

The EtD framework serves as a structured decision-making tool that can effectively bridge the strength of evidence with practical decision-making factors through multidisciplinary panel discussions.^45,46 Traditional EtD frameworks encompass a wide array of decisional criteria, such as the benefits and harms of an intervention, the certainty of evidence, resource expenditure, values, equity, and feasibility. Although these frameworks are comprehensive, they often prove to be complex and lengthy in the context of policy communication, hindering comprehension by non-experts and impeding swift action by decision-makers. As some research indicates, it is often neither necessary nor feasible to systematically evaluate all EtD domains in actual guideline development.⁴⁶ Therefore, to enhance the usability of scientific evidence, we have refined the traditional and often cumbersome EtD framework to construct a simplified EtD matrix, which retains only the most critical decision-making chain: risk–intervention–cost–policy.

Our analysis is structured around the following six key domains: Health Outcomes and Population at Risk, Pollutants and Effect Size, Certainty of Evidence, Recommended Intervention Plan, Cost Estimate/Cost Effectiveness, and Policy Priority and Basis for Determination. This approach aims to distill complex research into a transparent and actionable format for policy and public communication, thereby lowering the barrier to understanding and promoting rapid, consistent action focused on the most critical components in the field of drinking-water safety. By applying this EtD framework, we ensure a systematic and transparent evaluation process, providing clear and concrete scientific guidance for public health policy formulation.

Quantification and statistical analysis

During the data analysis phase, this study consistently applied a random-effects model (using restricted maximum likelihood [REML] method) to recalculate effect sizes for all included meta-analyses, ensuring uniformity in statistical methods. All effect sizes were log-transformed, with positive values explicitly representing increased risk due to exposure or intervention. Considering that most studies did not provide baseline incidence rates for exposed populations, and recognizing the potential biases introduced by forced conversions between RR and OR, this study refrained from converting between types of effect measures. Heterogeneity was assessed using the I² statistic, with 50% as the threshold for significant heterogeneity according to GRADE recommendations.⁴⁷ Evidence was downgraded accordingly in the GRADE rating when I² exceeded 50%. Publication bias was evaluated using Egger’s regression for meta-analyses containing 10 or more studies, with a p-value < 0.05 indicating significant publication bias.⁴⁸ Funnel plots were additionally employed to visually assess potential publication bias. Detailed results are provided in the supplemental information.⁴⁹

Robustness and prospective scenario analyses were conducted to comprehensively evaluate the reliability of meta-analytic findings and the need for further research. For statistically significant associations (p < 0.05), we used the classical Rosenthal Fail-safe Number (FSN) to assess robustness to publication bias. The Rosenthal FSN estimates, via sensitivity analysis, how many hypothetical unpublished studies with a null effect (i.e., effect size of zero) would be required to reduce a currently significant result (p < 0.05) to non-significance (p ≥ 0.05). Specifically, each original study’s effect size (y_i) and standard error (SE_i) were transformed to a standard normal statistic (Z = y_i/SE_i), the Z-values were summed to obtain the combined Z (∑Z), and, using the two-sided α = 0.05 critical value (Z_crit = ±1.96), the FSN was calculated as: $\text{FSN}={\left(\sum \mathrm{Z}∕{\mathrm{Z}}_{\text{crit}}\right)}^2-\mathrm{k}$, where k is the number of included primary studies. Following Rosenthal’s canonical criterion, when FSN > 5k + 10, the statistically significant association is considered highly robust to potential publication bias.⁵⁰

For associations that were not statistically significant (p ≥ 0.05), we performed prospective scenario analyses to determine whether additional research is warranted. Under a REML, we conducted computer simulations by sequentially adding hypothetical future studies and repeatedly re-estimating the pooled effect until nominal statistical significance (α = 0.05) was achieved. Two scenarios were specified: a conservative scenario assuming the true effect of future studies equals the current pooled effect (θ∗ = θ_pooled); and an optimistic scenario assuming the future effect equals the effect size of the currently most heavily weighted study (θ∗ = θ_max). In both scenarios, the variance of the future study effect sizes was assumed to equal the mean variance of the included studies. Beyond nominal significance, we also computed the scenarios required to reach 80% statistical power (CP80), i.e., the number of new studies needed such that the probability of detecting the assumed effect reaches 80%. If even a large number of additional studies cannot achieve significance or CP80, this signals the need to reassess the assumed effect size and weighting scheme.⁵¹

Published: February 4, 2026

Supplemental information

Data S1. Detailed characteristics and re-analysis results of meta-analyses on health risks, related to Figures 2 and 3

This spreadsheet contains the baseline characteristics, recalculated pooled effect sizes, and heterogeneity statistics for all included studies assessing the associations between unsafe drinking water and health outcomes.

Data S2. Detailed characteristics and re-analysis results of meta-analyses on intervention effectiveness, related to Figures 4 and 5

This spreadsheet provides the baseline characteristics, recalculated pooled effect sizes, and heterogeneity statistics for all included studies evaluating the effectiveness of drinking water quality interventions.

Data S3. GRADE evidence profiles and evidence classification for health risk associations, related to Figures 2 and 3

This file presents the details for the complete GRADE quality assessment domains for each intervention outcome.

Data S4. GRADE evidence profiles and evidence classification for intervention effectiveness, related to Figures 4 and 5

This file presents the details for the complete GRADE quality assessment domains for each intervention outcome.

Data S5. PRISMA checklist, related to STAR Methods

This checklist outlines the page numbers and sections where each item of the PRISMA reporting guidelines is addressed in the manuscript.