Soil-transmitted helminth infection, loss of education and cognitive impairment in school-aged children: A systematic review and meta-analysis

Background Evidence of an adverse influence of soil transmitted helminth (STH) infections on cognitive function and educational loss is equivocal. Prior meta-analyses have focused on randomized controlled trials only and have not sufficiently explored the potential for disparate influence of STH infection by cognitive domain. We re-examine the hypothesis that STH infection is associated with cognitive deficit and educational loss using data from all primary epidemiologic studies published between 1992 and 2016. Methods Medline, Biosis and Web of Science were searched for original studies published in the English language. Cognitive function was defined in four domains (learning, memory, reaction time and innate intelligence) and educational loss in two domains (attendance and scholastic achievement). Pooled effect across studies were calculated as standardized mean differences (SMD) to compare cognitive and educational measures for STH infected/non-dewormed children versus STH uninfected /dewormed children using Review Manager 5.3. Sub-group analyses were implemented by study design, risk of bias (ROB) and co-prevalence of Schistosoma species infection. Influential studies were excluded in sensitivity analysis to examine stability of pooled estimates. Findings We included 36 studies of 12,920 children. STH infected/non-dewormed children had small to moderate deficits in three domains—learning, memory and intelligence (SMD: -0.44 to -0.27, P<0.01–0.03) compared to STH-uninfected/dewormed children. There were no differences by infection/treatment status for reaction time, school attendance and scholastic achievement (SMD: -0.26 to -0.16, P = 0.06–0.19). Heterogeneity of the pooled effects in all six domains was high (P<0.01; I2 = 66–99%). Application of outlier treatment reduced heterogeneity in learning domain (P = 0.12; I2 = 33%) and strengthened STH-related associations in all domains but intelligence (SMD: -0.20, P = 0.09). Results varied by study design and ROB. Among experimental intervention studies, there was no association between STH treatment and educational loss/performance in tests of memory, reaction time and innate intelligence (SMD: -0.27 to 0.17, P = 0.18–0.69). Infection-related deficits in learning persisted within design/ROB levels (SMD: -0.37 to -52, P<0.01) except for pre-vs post intervention design (n = 3 studies, SMD = -0.43, P = 0.47). Deficits in memory, reaction time and innate intelligence persisted within observational studies (SMD: -0.23 to -0.38, all P<0.01) and high ROB strata (SMD:-0.37 to -0.83, P = 0.07 to <0.01). Further, in Schistosoma infection co-prevalent settings, associations were generally stronger and statistically robust for STH-related deficits in learning, memory and reaction time tests(SMD:-0.36 to -0.55, P = 0.003–0.02). STH-related deficits in school attendance and scholastic achievement was noted in low (SMD:-0.57, P = 0.05) and high ROB strata respectively. Interpretation We provide evidence of superior performance in five of six educational and cognitive domains assessed for STH uninfected/dewormed versus STH infected/not-dewormed school-aged children from helminth endemic regions. Cautious interpretation is warranted due to high ROB in some of the primary literature and high between study variability in most domains. Notwithstanding, this synthesis provides empirical support for a cognitive and educational benefit of deworming. The benefit of deworming will be enhanced by strategically employing, integrated interventions. Thus, multi-pronged inter-sectoral strategies that holistically address the environmental and structural roots of child cognitive impairment and educational loss in the developing world may be needed to fully realize the benefit of mass deworming programs.


Methods
Medline, Biosis and Web of Science were searched for original studies published in the English language. Cognitive function was defined in four domains (learning, memory, reaction time and innate intelligence) and educational loss in two domains (attendance and scholastic achievement). Pooled effect across studies were calculated as standardized mean differences (SMD) to compare cognitive and educational measures for STH infected/nondewormed children versus STH uninfected /dewormed children using Review Manager 5.3. Sub-group analyses were implemented by study design, risk of bias (ROB) and co-prevalence of Schistosoma species infection. Influential studies were excluded in sensitivity analysis to examine stability of pooled estimates.

Findings
We included 36 studies of 12,920 children. STH infected/non-dewormed children had small to moderate deficits in three domains-learning, memory and intelligence (SMD: -0.44 to -0. 27 differences by infection/treatment status for reaction time, school attendance and scholastic achievement (SMD: -0.26 to -0.16, P = 0.06-0. 19). Heterogeneity of the pooled effects in all six domains was high (P<0.01; I 2 = 66-99%). Application of outlier treatment reduced heterogeneity in learning domain (P = 0.12; I 2 = 33%) and strengthened STH-related associations in all domains but intelligence (SMD: -0.20, P = 0.09). Results varied by study design and ROB. Among experimental intervention studies, there was no association between STH treatment and educational loss/performance in tests of memory, reaction time and innate intelligence (SMD: -0.27 to 0.17, P = 0.18-0.69). Infection-related deficits in learning persisted within design/ROB levels (SMD: -0.37 to -52, P<0.01) except for pre-vs post intervention design (n = 3 studies, SMD = -0.43, P = 0.47). Deficits in memory, reaction time and innate intelligence persisted within observational studies (SMD: -0.23 to -0.38, all P<0.01) and high ROB strata (SMD:-0.37 to -0.83, P = 0.07 to <0.01). Further, in Schistosoma infection co-prevalent settings, associations were generally stronger and statistically robust for STH-related deficits in learning, memory and reaction time tests(SMD:-0.36 to -0.55, P = 0.003-0.02). STH-related deficits in school attendance and scholastic achievement was noted in low (SMD:-0.57, P = 0.05) and high ROB strata respectively.

Interpretation
We provide evidence of superior performance in five of six educational and cognitive domains assessed for STH uninfected/dewormed versus STH infected/not-dewormed school-aged children from helminth endemic regions. Cautious interpretation is warranted due to high ROB in some of the primary literature and high between study variability in most domains. Notwithstanding, this synthesis provides empirical support for a cognitive and educational benefit of deworming. The benefit of deworming will be enhanced by strategically employing, integrated interventions. Thus, multi-pronged inter-sectoral strategies that holistically address the environmental and structural roots of child cognitive impairment and educational loss in the developing world may be needed to fully realize the benefit of mass deworming programs.

Author summary
Previous systematic reviews of the effect of STH infection on cognitive and educational performance were either inconclusive or found little to no evidence of associated benefit. Lack of consensus in prior reviews and their narrow emphasis on randomized controlled trials prompted this comprehensive assessment of whether educational and cognitive benefits are associated with the common practice of preventive chemotherapy among children in STH-endemic regions. This literature synthesis included all relevant epidemiologic studies regardless of design and investigated associations between STH infections and two outcomes: i) educational loss (attendance and scholastic achievement) and ii) psychometrically performance in four neurocognitive domains-memory, learning, reaction time and intelligence. Pooled results across 36 observational and intervention studies suggest that STH infection/non-treatment is associated with deficits in five of six domains evaluated and provide evidence in support of a cognitive and educational benefit of deworming for STH infections. This overall finding was sensitive to study design and risk of bias across studies.

Introduction
Soil-transmitted helminthiases (STH) include infections with roundworm (Ascaris lumbricoides), whipworm (Trichuris trichiura) and hookworm (Ancylostoma duodenale and Necator americanus). STH infection affects one third of the world's population [1,2]. Primarily due to poverty, poor personal hygiene, frequent outdoor exposures, a higher likelihood of indulging in high risk behavior such as soil-eating, and the presence of environmental conditions favoring transmission [3,4], school-age children between the ages of 5 to 15 in mostly developing countries are at highest risk of chronic helminth infection and helminth-associated morbidities [2,4,5]. Of note, the designation of school-aged children as highest risk may be because there has been insufficient research of the neurodevelopmental and cognitive importance of STH in toddlers [6][7][8] and preschool children [6,9,10]. In this highly affected demographic, STH-related morbidities occur during critical periods of physiologic, mental and physical development. The chronicity of infection ensures that any STH-related small to moderate nutritional, growth and cognitive deficits is cumulated over extended periods of the developmental life course. Epidemiologic studies have reported lower health outcomes including-a higher prevalence of lethargy [11], stunting, wasting and anemia [12,13] for helminth-infected relative to uninfected school-age children. In addition, some epidemiologic studies have reported a higher prevalence of school-absenteeism [13], and lower performance on a range of cognitive tests [14]. The mechanisms of adverse effect on cognition and mental function of infected children are not well understood [15] but helminth-associated iron-deficiency is thought to be important [1,16,17]. Recent evidence suggests that helminth-associated morbidity and mortality is likely magnified for polyparasitized children [18][19][20] and those with chronic untreated moderate and heavy intensity of infections.
Recent systematic reviews and meta-analysis of the health benefits of STH treatment based on intervention study designs concluded that the evidence for improvements in cognitive function was mixed and inconclusive and based on at best low quality evidence [15,21]. Conclusions from these meta-analyses raise fundamental questions about the expected educational and cognitive benefits that partly justify current deworming programs in helminth endemic regions. However, the following criticism have been made of their conclusions [22][23][24][25][26]: (i) existing trials included were of poor quality [25]; (ii) interpretation did not factor in realities such as the high frequency and rapidity of reinfection following treatment [22]; (iii) and the fact that by offering one time treatment combined with the short-term duration of most trials, they have limited ability to evaluate long-term benefits of regular deworming as envisioned and supported by the World Health Organization [5], Unesco and the World Bank [23]. Proper assessment of long-term cognitive outcomes may require a sustained period of intervention and possibly post-treatment educational remediation in school-aged children [27][28][29][30][31][32].
Given the limitations of randomized trials noted above and a health policy environment that supports periodic deworming of school-aged children for expected growth, nutritional and perhaps cognitive benefits, clinical equipoise for unknown benefit of deworming is difficult to justify and the ethical landscape has coalesced around the understanding that randomizing children to deworming compared to placebo is largely unjust. Systematic reviews and meta-analyses remain an important tool for clarifying the possible cognitive impact of deworming. Including observational and intervention study designs in this synthesis provides robust scientific evidence to inform existing empirical gap in a way that enhances external validity beyond highly selected trial participants and conforms with the prevailing clinical, ethical and health policy environments that converge and contend with one another on the subject of deworming.
Hence, we incorporate all available epidemiologic evidence and re-examine the hypothesis that helminth infections adversely affect child cognitive function and educational outcomes. Unlike prior reviews, we examine the impact of STH infection on specific cognitive domainslearning, memory, attention/reaction time and intelligence, and do not assume equal impact of STH on across domains. We specifically hypothesize that STH infections will be associated with educational deficits and lower performance in learning, memory and attentional processes because these domains of child function are more sensitive to environmental assaults. We define a fourth domain "innate intelligence" that we expect to be mostly genetically determined and possibly less affected by STH infection.

Selection of studies
We performed a comprehensive search of databases including MEDLINE using PubMed (http://www.ncbi.nlm.nih.gov), Science Direct and Google Scholar, as of April 22, 2016 unrestricted by language. Throughout the data search, we used the key words, "soil-transmitted helminthiasis" and "cognitive" as well as "intestinal parasites", "school performance" and "school attendance". We included all studies-regardless of study design that had raw data for relationships between any of the 3 STH parasite species and educational indices (attendance or achievement) or neurocognitive outcomes. Articles were excluded if: (i) they were duplicates; (ii) completely lacked or incompletely presented needed raw data for relationships between STH and respective outcomes. We searched the body of included studies and their reference sections [32][33][34][35] for possible additional inclusions to this meta-analysis not identified by keyword searches. Our search yielded 7,709 citations and after a series of exclusions (outcomes attributed to other than STH, absence of required data and presence of duplicate data), provided for the final pool of articles included in the meta-analysis which was [13,17,31,. We focused on the more recent publications (from the year 1992 onwards) in the hope of minimizing the high level of heterogeneity across studies prior to 1992 as noted by Watkins and Pollitt. [35].

Psychometrically assessed cognitive functions
These were categorized into four domains: (i) memory, (ii) reaction time, (iii) learning, and (iv) intelligence tests. Many studies used a suite of psychometric instruments to assess a single or multiple cognitive domains in enrolled children. The memory domain included tests of working (short-term) memory as well as those of long-term memory. Attention/reaction time tests were those that measured the ability of a child to sustain concentration on a particular object, action, or thought, including their capacity to manage competing demands in their environments. The learning/executive function domain included tests to evaluate children's performance in goal-oriented behavior, particularly components that are important for scholastic advancement. Executive function included tests of cognitive processes that enable children to connect past experience with present action, and by so doing, engage in planning, organizing, strategizing, paying of attention to details, and to emotionally self-regulate, make necessary efforts to remember important details required for attainment of future goals. [69] We included in the 'intelligence' domain psychometric tests of intelligence quotient (IQ) that we believe measures largely biologically determined cognitive abilities, in contrast to cognitive performance measures that are environmentally pliable. [70] When multiple instruments were used to measure the same cognitive domain, a grand mean of scores and a grand mean of standard deviation (SD) across all instruments were calculated. Thus, for each publication, one overall mean and SD value was determined for each domain. A study could contribute data to different cognitive domains if it used tools spanning across several cognitive domains; however, each instrument only contributed to one single domain of function (S1 Table). Overall effects were derived only if there were two or more studies in a given domain.

Educational deficits
Attendance rate was defined as the number of days children attended school over the past month (in cross-sectional studies) or over the study period (in longitudinal studies). In casecontrol studies, the percentage of children enrolled versus not enrolled in school was calculated for STH-infected and non-infected children.
The definition of scholastic achievement varied across studies. Children with high versus low attainment were identified based on: i) pass rate on standardized teacher-made tests, ii) percent of children that were in appropriate class for their age, iii) the enrollment of children in elite versus non-elite schools, iv) scores in the school function domain of pediatric qualityof-life inventory, v) change in class position after treatment for STH infection, vi) an above versus below average class performance as rated by a teacher, or vii) pass rate in any kind of educational test, whether teacher administered or not. (S1 Table).

Data extraction
Two investigators (NP, AEE) independently extracted data. Disagreements (if any) were resolved by a third person (LT). If resolution was not attained by the third author alone, abstracted information was resolved by consensus between LT, NP and AEE. The following information was obtained from each publication: first author's name, published year, country of origin, parasite(s) involved, age range of the subjects, effect outcome of the study, study design, study features as well as subject features and sample sizes.

Quality assessment of the included studies
We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guidelines in describing our findings and standard methodology [71][72][73]. We assessed study quality using the Cochrane Collaborations tool for assessing risk of bias in randomized trials. [74] To assess methodological quality of each observational study, we modified the Newcastle-Ottawa scale (NOS) [75,76] using the star Ã system based on the following: (i) representativeness of infected sample or case/control selection (min = 0, max = 3 Ã ); (ii) comparability using known correlates of cognitive function/educational attainment (min = 0, max = 6 Ã ). Here, we accounted for the confounding effects of age (score = 1 Ã ), sex (score = 1 Ã ), nutritional (score = 2 Ã ) and socioeconomic (score = 2 Ã ) status in relation to STH infection and educational/ cognitive outcomes. (iii) Absence of bias in relation to outcome assessment in prospective cohort studies (min = 0, max = 3 Ã ) or exposure assessment in cross-sectional and case-control studies (min = 0, max = 3 Ã ). For each study, the initial raw quality score (max = 12 Ã ) was rescaled to match the scale of 9 Ã and then classified as low, high or very high risk of bias per prior literature precedent [76].

Statistical analysis
Data were analyzed using Review Manager 5.3 (Copenhagen: Nordic Cochrane Centre, Cochrane Collaboration, 2014) and SigmaStat 2.3 and SigmaPlot 11.0 (Systat Software, San Jose, CA). We sought the number of children on two levels: (i) those who took the cognitive tests and compared those who were infected with those who were not; and (ii) those monitored for school achievement and attendance. We expressed these relationships as standardized mean difference (SMD) and 95% confidence intervals (CI). SMD estimates were classified as statistically significantly different if their confidence intervals did not cross zero. Most the included articles (studies) presented multiple means and measures of spread (SD or standard error (SE)). From each article, we obtained the mean of the multiple means and SDs which we used as input in generating forest plots. However, Some papers presented median (m) and range (a and b). These measures were converted into approximate mean and SD following Hozo et al [77]. Where mean and 95% CI or SE were reported, SD were derived as described by the Cochrane Collaboration [78]. For studies presenting data on differences in mean scores between two time points for treated/infected versus untreated/uninfected groups, appropriate SD for mean difference was calculated per the approach described by the Cochrane Collaboration [78]. Pooled SMDs were obtained using two analysis models based on presence or absence of heterogeneity: the fixed [79] and random [80] effects, respectively. Heterogeneity refers to diversity which may influence the manner in which the data are treated [81]. Significance was set at a P-value of <0.05. We addressed heterogeneity between studies with the following: (i) estimated using the χ 2 -based Q test [82]; (ii) its sources identified using the Galbraith plot method [83] and (iii) explored using subgroup analysis [82] where we examined effects in observational and interventional studies. Intervention studies, often prospective, are specifically tailored to evaluate direct impacts of treatment or preventive measures on disease. Observational studies on the other hand are often retrospective and used to assess potential causation in exposure-outcome relationships and therefore influence preventive methods [84]. In addition, we also examined effects according to risk of bias (low and high), all trial and prepost studies as well Schistosomiasis coinfection.
Influence of each study on robustness of the summary effects and heterogeneity was determined with sensitivity analysis, which involved omitting one study at a time followed by recalculation of the pooled SMD. Change in direction of association (e.g. poor performance to better performance and vice-versa) after study omission indicates non-robustness of the summary effect, otherwise the pooled SMD is considered robust, indicating stability of the results. We investigated publication bias in domains with ! 10 studies because with < 10, sensitivity of the qualitative and quantitative tests of comparisons studies was low [85]. We implemented sensitivity analysis based on observational or experimental study design. Available data from studies that provided baseline treatment of children and post-test readings of the same subjects were included as intervention studies.

Included studies
We included 36 studies (Fig 1) of 12,920 children (5-20y) that evaluated STH associated differences in psychometrically evaluated cognitive tests, educational attainment or school attendance (Table 1). Of these, 5,932 children were evaluated in the context of STH treatment (with or without randomization) and 6,978 were evaluated as part of observational study. In the latter, 5,538 and 586 children were part of cross-sectional and longitudinal studies, respectively. The remaining 854 subjects were part of comparative Epidemiologic surveys. The PRISMA checklist is included with further pertinent details for this meta-analysis (S3 Table). Table 2 ranks included observational studies in terms of methodological quality using the NOS. Fourteen (56%) of these were determined to have low risk of bias and 11 studies (44%) were classified as having high risk or very high of bias. The ten intervention studies with featuring allocation of trial participants to deworming (alone or in combination) versus control groups were communicated across 11 publications. Seventy percent of trials had moderate or high risk of bias and thirty percent were uncertain to low risk of bias ( Table 3).

STH species and treatment effects
S2 Table shows that the vast majority of included studies examined educational or cognitive impact infection with multiple STH species (up to 62.5%). Ascaris infections were the least studied (none in three domains). Nineteen studies (52.7%) included in this meta-analysis used Albendazole or Mebendazole for STH treatment with the exception of Boivin et al [41] where Decaris (Tetramisole, Lavamisole) was used.

Impact of STH infection on educational loss and cognitive functions
As shown in Table 4, the overall analysis shows that STH-infected/non-treated children performed significantly worse than uninfected/dewormed children in the three of the six domains    including: memory (SMD: -0.31, P = 0.01); learning (SMD: -0.44, P <0.0001, Fig 2) and intelligence (SMD: -0.27, P = 0.03). Because pooled effects in all six domains were obtained under highly heterogeneous conditions (P heterogeneity <0.01, I 2 = 66-99%), we identified investigations contributing to large variability in the pooled effect of STH-infection/non-treatment on performance in each of the cognitive and educational domains using the Galbraith plot technique (Fig 3). This technique did not significantly reduce heterogeneity in respective outcomes with the exception of the learning domain (I 2 = 33%, Fig 4). However, application of this technique resulted in emergence of statistically robust estimates of effect in three domains: reaction time (SMD: -0.21, P = 0.004), achievement (SMD: -0.24, P <0.01) and attendance (SMD: -0.52, P <0.01). On the other hand, exclusion of one influential study (Shang et al [61]) resulted in loss of statistical significance in intelligence domain (SMD: -0.20, P = 0.09).

Subgroup analysis
Stability of overall findings were evaluated in sub-groups distinguished by study design, presence of non-outlier studies and study quality (Table 5). In the context of observational study designs, STH infection/non-treatment was significantly and consistently associated with deficits in memory, learning, reaction time and intelligence with SMD ranging from -0.42 to -0.23 (P <0.01-0.05). Among studies featuring STH treatment-pre-post longitudinal study or experimental in design, moderate (SMD: -0.8, P = 0.14) to large (SMD: -0.46, P<0.0001) deficits in memory and learning tests were respectively noted. was evident for untreated children.  Table 4 P<0.001) with corresponding substantial reduction in heterogeneity (P heterogeneity = 0.72, I 2 = 0%). Among studies classified as low risk of bias, STH infection/non-treatment was associated with significant deficits in memory, learning and school attendance (SMD: -0.39 to -0.55, P <0.0001-0.03). Among studies classified as high or very high risk of bias statistical significance was achieved or maintained in the domains of learning, reaction time and intelligence (SMD: -0.28 to -1.00, P <0.001-0.03) as well as achievement (SMD: -0.70 P <0.001).

Test of Association Test of Heterogeneity
To determine whether Schistosoma species co-infection to contributed to observed associations between respective outcomes and STH infection vs. no infection among observational studies, separate pooled estimates were derived for primary studies with and without co-prevalence of Schistosoma species (Table 6). STH infection related deficits in learning (SMD = -0.32, P = 0.02) and memory (SMD = -0.32, p = 0.001) was evident in pooled analyses of primary studies without co-prevalent Schistosoma species infection. Among observational studies with co-prevalent schistosoma co-infection, magnitudes of association were generally stronger and pooled associations were statistically for infection related deficits in tests of learning, memory and reaction time (SMD -0.36 to -0.55, P = 0.003-0.02). Table 7 presents a summary of the impact of deleting individual publications on robustness of pooled SMD in overall analyses and within strata of design, outlier study status and study quality. Of the six domains, reaction time was most robust followed by learning. In most subgroup analyses (post-outlier, observational, high risk of bias and those without Schistosomiasis coinfection) pooled SMDs were similar in magnitude and direction (i.e. robust) in comparison with SMDs from overall analyses. Of the 47 robust estimates, 27 (62.8%) provided statistically robust associations between STH infection/non-treatment and cognitive deficits or educational loss. We found no evidence of publication bias in all six outcome domains for the overall findings and subgroups with ! 10 studies ( Table 8).

Principal findings and interpretations
In line with our hypothesis, we found that non-treatment for STH infection was consistently associated with statistically, clinically and health policy relevant deficits in five of the six domains examined. The relationships observed were generally invariant to exclusion of influential individual studies but robustness of findings varied by study design-observational, STH treatment studies and experimental studies, and study quality. Specifically, infection related deficits were larger in magnitude, tended to be more statistically robust and stable in direction for pooled studies of observational compared to STH treatment (with or without experimental) studies and for high vs. low risk of bias studies. The direction of pooled effects from observational studies with and without potential for Schistosoma species coinfection were similar although the magnitude of associations tended to be higher for STH-related deficits in settings with potential for Schistosoma species coinfection. These non-treatment/STH-infection related disadvantages in cognitive function and educational measures correspond to small to moderate deficits per the Cohen criteria [88]. Of note, these small to moderate deficits are for a highly prevalent exposure affecting millions of children in the developing world. Because reinfection is rapid in helminth endemic settings [45], many children are effectively chronically infected by these parasites and the majority of children are simultaneously polyparasitized-i.e. infected by two or more STH species at the same time [19]. Polyparasitic and multi-species infections may produce additive disadvantages for cognitive function and educational loss. Our finding of stronger pooled effects in Schistosoma prevalent settings is suggestive evidence in support of this thesis. Hence, we speculate that the modest deficits reported here may under-estimate the true magnitude of learning and memory deficits attributable to untreated STH infections in endemic regions.
Working memory and learning tasks involve the brain's frontal lobe [89] which continues to develop throughout school-age years [90]. Drake et al in prior systematic reviews [18,32] noted the lack of consistency in cognitive domains affected by helminth infections. The one exception was in STH-infection associated lower performance on tests of working memory [32]. Our review lends credence to that previous observation for working memory and provides new evidence of STH infection related deficits in learning tests. Also in line with our study hypothesis, we found that untreated STH infection was associated with slower reaction time. The association between STH infection/non-treatment and attentional processes/reaction time was evident in observational studies and persisted among intervention studies when two influential intervention studies were excluded. These suggest that STH infections may have an adverse effect on attention and/or cognitive speediness (i.e. reaction time).
Two findings from this meta-analyses were either not consistent with our study hypothesis or were altogether surprising. The finding that school attendance and scholastic achievement did not differ significantly in overall analyses by infection/treatment status, though similar to results from prior meta-analyses on this subject, [21,34,[91][92][93] was contrary to our study hypothesis. However, we noted that the overall findings were not stable. With exclusion of influential studies and design stratified analyses we noted suggestive evidence of an adverse impact of infection/non-treatment on these outcomes among observational studies with high potential for bias but not among intervention or low risk of bias studies. Of note, scholastic . Squares indicate the SMD in each study, with square sizes directly proportional to the weight contribution (%) of each study. Horizontal lines represent 95% confidence intervals (CI). The Z test for overall effect indicates non-significance given that P > 0.05. The chi-square test indicates that heterogeneity (P = 0.12, I 2 = 33%) was reduced by outlier treatment. SD: standard deviation; CI: confidence interval; Std: standard; df: degree of freedom; I 2 : measure of variability expressed in %.
https://doi.org/10.1371/journal.pntd.0005523.g004 Table 5. Effect of STH infection/non-treatment on performance in cognitive and educational within strata of study designs, non-outlier investigations and study quality.

Test of Association
Test of Heterogeneity achievement, as operationally defined in this study, includes children's pass rate on standardized or teacher-made tests, the percent of children that were in appropriate class for age, child enrollment in elite versus non-elite schools, scores in the school function domain of pediatric quality-of-life inventory, change in class position after treatment, an above average versus average/below average teacher rating of scholastic performance and pass/fail rate in any kind of educational test, whether teacher administered or not. Hence, our finding of significant adverse effect of STH infection on this domain could have broader implications for the educational achievement in complex real-world settings. The effect on achievement likely reflect to varying degrees deficits in cognitive domains such as learning, memory and reaction time [94]. The observation of infection related deficits for tests of innate intelligence was surprising because we conceptually conceived innate intelligence as mostly a "genetically determined finite quantity" relatively insensitive to postnatal environmental assaults such as STH infections. Of note, this finding was robust only among observational studies and those classified as "high risk of bias". Furthermore, the impact of infection/non-treatment for STH on innate intelligence became statistically insignificant with the exclusion of an influential study suggesting the need for abundant caution in the interpretation of this data. Here, we have classified neurocognitive performance assessed with variety of instruments in the primary literature into four major cognitive domains based on extant knowledge of the primary cognitive domains covered by respective tools. We supplemented as needed with content area guidance from a neuropsychologist. Innate intelligence included measures of general intelligence based on IQ tests such as the Philippine non-verbal intelligence test, Kaufmann assessment battery for children, Wechsler intelligence test for children (where overall scores and no subscales are specified). Prior reviews [32] have eloquently described the inherent challenges with neurocognitive assessments that complicates empirical efforts to understand the impact of helminth infections on disparate domains. For example, performance in respective domains likely correlate with one another to varying degrees in most individuals. Hence, it is possible that the STH-associated deficits in innate intelligence measures partly reflect STH-related effect on other domains where performance may be more sensitive to environmental factors including educational quality and health factors [95]. For this systematic review and meta-analysis, we have intentionally taken the approach of combining assessment instruments and educational measures into one of six pre-identified domains that contribute in varying degrees to the ability of children to learn, be educated and potentially their future productivity as adults. These classifications are guided by the literature and underlying theoretic constructs assessed. In spite of our best efforts, this remains an imprecise science. We provided extensive detail of our classification approach as supplemental material for critical evaluation and further refinement. In sum, the high heterogeneity [96], limited statistical power in sub-group analyses [95,97] and the potential for residual confounding-particularly in non-randomized interventions and observational studies, [98] should lead to cautious interpretation of findings from sub-group and sensitivity analyses.

Limitations and strengths
Our review is subject to several limitations that should be considered in the interpretation of our findings. Firstly, heterogeneity was high across studies. We analytically addressed this using random-effects model and explored the specific role of influential studies using the Galbraith plot method. Secondly, as noted earlier, our overall findings were sensitive to observational versus intervention study design and study quality. Thirdly, we are unable to fully delineate potentially STH species-specific differences in the domains evaluated as studies did not always distinguish between STH species. Additionally, across individual studies, the number of stool samples used for diagnosis of STH infection varied from single to multiple. Where single samples were tested, the chance of misclassifying lightly infected children increases. Thus in some studies, "the uninfected" may include an unknown number of mostly lightly infected children. Similarly, information on joint Schistosoma and STH infections as well as data on STH infection intensity was not consistently provided making it difficult to robustly examine potential for beyond additive adverse effects for multi-species infected children and dose-response by infection intensity. Last but not least, variations in the primary literature did not allow for subgroup analyses by the following factors-STH species, malnourished versus normally nourished children, type of deworming agent, and treatment strategy (mass deworming versus test and treat only infected children). These subgroup analyses-if possible, could give insight with respect to the kinds of children most likely to derive a cognitive benefit from deworming and whether mass or targeted deworming is the best approach [99] to deworming school-aged and potentially preschool children [100]. Potential for heterogeneity in the association of deworming with cognitive and educational outcomes by child nutritional status, treatment strategy, deworming agent and parasite species requires further elucidation. In light of considerable heterogeneity, varying quality of underlying studies and sensitivity to experimental versus observational study design in some outcomes, we agree with prior expressed need for appropriate caution in the interpretation of findings from synthesis of epidemiologic literature especially where associations are primarily driven by observational studies and the risk of bias in underlying studies is high or very high [101]. In spite of these limitations, our study has a number of important strengths that allow us provide additional context for understanding the influence of study design, specific influential studies and risk of bias within present literature and STH-associated cognitive and educational deficits. We address two issues noted by the recent Campbell review-i.e. the use of a complicated array of different cognitive tools and a lack of understanding of STH impact on absenteeism [101]. We enhance clarity regarding the possible impact of STH infection/non-treatment on educational and cognitive loss by systematically collapsing the variety of cognitive tools used within four cognitive domains and conducting specific analyses on educational lossincluding scholastic achievement and absenteeism. Our approach implicitly recognizes STHinfections may have unequal impact in different domains despite potential over-lap in abilities tested within respective domains. It is noteworthy that in spite of major differences in our empirical approach relative to prior systematic reviews and meta-analyses on this subject, our finding of no STH-infection related deficits for most cognitive and educational outcomes in sub-group analyses restricted to experimental study designs confirms previously reported findings based on prior systematic reviews focused on RCTs only. [21,34,[91][92][93] This meta-analysis allows for evaluation of consistency in primary relations investigated across key factors by including all relevant epidemiologic studies regardless of design, conducting explicit analyses to evaluate impact of influential studies, and evaluating the risk of bias in underlying studies. It has been noted that carefully implemented meta-analyses based on observational studies generally produce estimates similar to those from meta-analyses based on RCT and thus supports evidence-based medical decision-making [102,103]. Our approach provides empirical evidence to evaluate an important health policy and yet allows allows for appropriate qualification as warranted on the basis of design, influential studies and risk of bias.

Conclusion, remaining gaps and recommendations for future research
We provide evidence of small to modest deficit in five of six evaluated domains although there were few influential studies and variations for associations existed by study design. Prior metaanalyses of RCTs that evaluated cognitive impacts of STH infections had different conclusions about the cognitive and scholastic effects of STH infection [15,21,34]. Key differences in our approach-namely inclusion of all relevant epidemiologic studies and evaluation of effects within educational and cognitive domains, may partly account for the inferential difference between this and prior reviews.
Despite the empirical debate regarding the cognitive benefit of deworming for STH, the current ethical, clinical and health policy environments remain strongly skewed in favor of deworming for child growth, prevention of anemia and potentially avoidance of preventable cognitive deficits. Deworming efforts have expanded as a strategy to control the prevalence and intensity of infections. Ongoing operational intervention research is poised to provide valuable insight regarding the optimal approach-whether school based, community based or a combination of infection control via mass deworming will additionally examine the net benefit of bi-annual versus annual deworming [104]. Mass deworming campaigns are unlikely to interrupt infection in most settings, but infection profile is expected to shift towards low-intensity single and multi-species co-infections. Prior research has demonstrated that polyparasitism is consequential for anemia-an important determinant of cognitive deficit in children [19,105]. Specific investigations of the cumulative impact of polyparasitic STH infection on cognitive and educational outcomes are lacking. Future epidemiologic studies that explicitly examine the cognitive impact of multi-species parasitic infections by number as well as by intensity of coincident infections will provide useful information to guide health policy and may inform the optimal frequency of deworming in the context of low intensity infections.
Currently, preschool-age children are not treated as part of routine deworming programs for STH [106], and yet, the evidence suggests that children are infected shortly after weaning and remain persistently infected throughout childhood and adolescence. The adverse impact of infection on children begins way before school age and compounds the cumulative health disadvantage associated with STH infection. This meta-analysis, and indeed most short-term study outcomes meta-analyzed, does not include this demographic. Recent review of investigations have demonstrated the safety and efficacy of STH deworming in preschool children [107]. Existence of an adverse developmental impact of STH infection on cognitive/educational domains would justify expanding the age-bracket of children who should be subjected to STH deworming. Educational and cognitive interventions will likely be more effective if initiated earlier in life for STH-infected children. However, more investigations among preschool children may be needed to understand the risks and potential benefits of early deworming in these children.

Policy implications
Given the chronicity of infection during childhood and adolescence in helminth endemic regions, it is possible that any cognitive and educational loss associated with STH infection will not be resolved by deworming alone without prevention of reinfection and specific interventions to address the environmental and structural determinants of parasitic infections. Thus, effective future interventions are likely to be those that emphasize multi-pronged inter-sectoral strategies to holistically address challenges to child wellbeing in the developing world [108].
Supporting information S1  . Squares indicate the SMD in each study, with square sizes directly proportional to the weight contribution (%) of each study. Horizontal lines represent 95% confidence intervals (CI). This treatment generated gain in significance (P = 0.004) for overall effect but did not affect heterogeneity (P <0.00001, I 2 = 78%). SD: standard deviation; CI: confidence interval; Std: standard; df: degree of freedom; I 2 : measure of variability expressed in %. . Squares indicate the SMD in each study, with square sizes directly proportional to the weight contribution (%) of each study. Horizontal lines represent 95% confidence intervals (CI). The Z test for overall effect indicates significance (P < 0.05) and the chi-square test indicates that heterogeneity (P <0.00001, I 2 = 80%). SD: standard deviation; CI: confidence interval; Std: standard; df: degree of freedom; I 2 : measure of variability expressed in %. . Squares indicate the SMD in each study, with square sizes directly proportional to the weight contribution (%) of each study. Horizontal lines represent 95% confidence intervals (CI). This treatment resulted in loss of significance (P = 0.09) for overall effect but did not affect heterogeneity (P = 0.00001, I 2 = 77%). SD: standard deviation; CI: confidence interval; Std: standard; df: degree of freedom; I 2 : measure of variability expressed in %. . Squares indicate the SMD in each study, with square sizes directly proportional to the weight contribution (%) of each study. Horizontal lines represent 95% confidence intervals (CI). The Z test for overall effect is not significant (P > 0.05) and the chi-square test indicates presence of heterogeneity (P <0.00001, I 2 = 93%). SD: standard deviation; CI: confidence interval; Std: standard; df: degree of freedom; I 2 : measure of variability expressed in %. (TIF) S10 Fig. Effects of outlier treatment on the overall results in the achievement domain. Diamond denotes the pooled standardized mean difference (SMD). Squares indicate the SMD in each study, with square sizes directly proportional to the weight contribution (%) of each study. Horizontal lines represent 95% confidence intervals (CI). This treatment generated gain in significance (P = 0.008) but did not affect heterogeneity (P = 0.00001, I 2 = 90%). SD: standard deviation; CI: confidence interval; Std: standard; df: degree of freedom; I 2 : measure of variability expressed in %. (TIF) S11 Fig. Forest plot of the overall results in the attendance domain. Diamond denotes the pooled standardized mean difference (SMD). Squares indicate the SMD in each study, with square sizes directly proportional to the weight contribution (%) of each study. Horizontal lines represent 95% confidence intervals (CI). The Z test for overall effect is not significant (P > 0.05) and the chi-square test indicates presence of heterogeneity (P <0.00001, I 2 = 99%). SD: standard deviation; CI: confidence interval; Std: standard; df: degree of freedom; I 2 : measure of variability expressed in %. (TIF) S12 Fig. Effects of outlier treatment on the overall results in the attendance domain. Diamond denotes the pooled standardized mean difference (SMD). Squares indicate the SMD in each study, with square sizes directly proportional to the weight contribution (%) of each study. Horizontal lines represent 95% confidence intervals (CI). This treatment generated gain in significance (P = 0.008) but did not affect heterogeneity (P = 0.00001, I 2 = 98%). SD: standard deviation; CI: confidence interval; Std: standard; df: degree of freedom; I 2 : measure of variability expressed in %. (TIF)