2.1 Eligibility criteria

We included peer-reviewed studies and published theses reporting on trachoma prevalence (cross-sectional, case-control, and cohort study designs) that described associations between trachoma prevalence and climatic, demographic, environmental, infrastructural, socioeconomic, geographic, and spatial proximity factors measured at community level. Full Inclusion and exclusion criteria are listed in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Inclusion and exclusion criteria used for selecting studies for the full review.

https://doi.org/10.1371/journal.pntd.0010272.t001

2.2 Information sources and search strategy

We conducted the systematic search in five databases: Medline (US Medical Library/Ovid), Embase (Elsevier/Ovid), Global Health (CABI, Ovid), Dissertations & Theses Global, and Web of Science (Clarivate Analytics). Two search rounds were conducted, by the lead researcher (Burgert), in all five databases, first on 2 July 2020 for the dates January 1950 through June 2020 and later updated on 3 February 2021 to include July 2020 through January 2021. The search terms were a mix of medical subject header terms (MeSH) and free text keywords; the specific terms used were adapted for each database. The final search terms for each database are summarised in S3 Text.

The basic structure of the search was split into two main components joined by AND:

• Component 1: trachoma terms (e.g., trachoma, trichiasis, ocular chlamydia) joined by OR
• Component 2: associated factor terms joined by OR
  1. ○ Climatic terms (e.g., rain, temperature, sunshine, aridity)
  2. ○ Demographic terms (e.g., population density, crowding, migration)
  3. ○ Environmental terms (e.g., altitude, soil, land cover)
  4. ○ Infrastructure terms (e.g., rural/remoteness, access, water, sanitation)
  5. ○ Socioeconomic terms (e.g., housing, crowding, population, poverty)
  6. ○ Geographic/spatial terms (e.g., GIS, raster, spatial clustering)

2.3 Selection process

All records were assembled, and duplicates were removed using Endnote Version X8.2. All retained non-duplicate studies were reviewed in a two-step process. Each study title/abstract was reviewed in step one and given a primary reason for exclusion/inclusion by two researchers working independently (Burgert and either Adams or Mingkwan), recorded in a Microsoft Excel Version 2012 [24] table. The reasons for exclusion were:

1. Missing abstract or full text not found
2. Language other than English, French, Spanish, or Portuguese
3. Non-human research (genomic, cellular, or other laboratory-based)
4. Systematic review, guidelines, mathematical modelling study, or opinion piece
5. Study of condition(s) not related to the eyes, including sexually transmitted infections
6. Report on trachoma research not including prevalence results, such as MDA coverage or impact of treatment on other disease outcomes only
7. Report on individual or household trachoma prevalence alone, only individual- or household- associated risk factors or no geospatial methods used

In step two, the abstracts of all studies meeting the population, outcome, type of study, date, and language criteria, regardless of the inclusion of selected interventions (exclusion criteria 6 and 7 plus all included studies), were reviewed by three researchers independently (Burgert, Adams, and Mingkwan). If any reviewer felt that any abstract met the inclusion criteria, the study was included in the next stage. A full-text review of studies for final inclusion was performed by two researchers independently (Burgert and either Adams or Mingkwan). The three research team members discussed any disagreements, and a consensus agreement was made regarding inclusion. Furthermore, studies were excluded if an overall relationship/trend (e.g., percentage, odds ratio [OR]) could not be synthesised between the associated factor described and the outcome.

2.4 Data extraction

The extraction of information for included studies was performed by one researcher (Burgert) and reviewed by one researcher (Adams). Extracted data were tabulated in Microsoft Excel Version 2012 [24]. The primary data items collected for this review were trachoma outcome measures, associated factor variables, how factors were measured, and data source of factors; along with study year, population, sampling, and design. The primary outcome domains recorded were the signs of TF and TT from the WHO simplified grading system [2,3] or ocular Ct infection (measured by a nucleic acid amplification-based test). Though the WHO simplified grading system was not officially published until 1987, studies measuring the primary outcomes of TF or TT using similar or comparable grading methods prior to 1987 were included [2]. Trachoma outcomes were recorded as measured in the study, such as TF or TT, including any specific age ranges. Analysis type, univariable (outcome and associated factor only) or multivariable (outcomes and associated factor plus other confounding factors such as age and sex/gender), were recorded for all studies.

Specific associated factor variables measured in the studies were combined into variable groups, e.g., precipitation (variable group) encompasses mean annual precipitation (specific variable) and monthly rain days (specific variable). Variable groups were further combined into variable categories, e.g., climate (variable category) includes temperature (variable group) and precipitation (variable group). Five variable categories were used: climate, demography, environment, infrastructure, and socioeconomic status. For studies that included multiple outcome measures or multiple associated factors, an individual row for each combination of the outcome, associated factors, and analysis type was included in the Excel spreadsheet. For example, studies reporting both a univariable and multivariable result for an outcome and associated factor pair had both results recorded. Thus, a paper reporting univariable and multivariable results on TF and TT with associated factors of altitude and temperature would have a total of eight rows in the spreadsheet. The data source for each associated factor was also recorded, including if it was an external geospatial raster, an external georeferenced data source, or collected/measured within the same survey as the trachoma outcome data. Study results were recorded as ORs or percentages, along with any reported confidence intervals (CI) and p-values, and any controlling factors from the analysis. The statistical significance of results, either noted as p-values or non-overlapping confidence intervals, was also documented.

2.5 Quality assessment

Study quality was assessed using the STROBE checklist for observational studies (cohort, case-control, and cross-sectional studies) [25]. Specific areas considered to assess overall quality were:

1. Survey type: population-based, cohort, or school-based.
2. Survey population and level of representation: population target, population examined, number of survey clusters, and number of evaluation units/districts or other levels of representation for the sample.
3. Use of WHO survey guidance [26,27] or similar for survey sampling methods; although many studies might have taken place earlier than the guidance was released, all studies were measured against the current WHO guidance.
4. Survey focus: trachoma only or multiple diseases/outcome measures.
5. Method of trachoma identification: using WHO simplified grading system to measure trachoma outcomes [2,3] and ophthalmic magnifying loupes.

An overall quality assessment was given for each study. A high assessment was given to studies that met the current WHO guidance, including being population-based, use of the correct study population for outcome of interest, inclusion of 20–30 survey clusters per evaluation unit, and positioning of trachoma as a primary outcome of the survey. A medium assessment was given to studies that met most but not all the criteria for a high assessment, while a low-quality assessment was for cohort or school-based studies or those in which the population-based sample was not sufficient to be representative of a given evaluation unit.

2.6 Data synthesis and certainty assessment methods

Due to the heterogeneity of study methods and differences in how associated factors were reported (varying results categories, e.g., altitude measured as a continuous or categorical variable), a meta-analysis was not possible and instead a synthesis of results was created. Factors associated with the outcome were summarised to indicate trend direction (e.g., increasing altitude) and impact on outcome (e.g., lower TF prevalence). For factors that were categorical by nature, such as urban versus rural areas, one category was selected and assessed by comparing how its presence/absence impacted the outcome (e.g., higher TF prevalence).

Table 2 describes the three results categories and the types of results recorded in each. Studies with both univariable and multivariable results for the same outcome and associated factor pair were combined to report only one result for the included study. If the univariable and multivariable results differed within the same study, the outcome was listed as "mixed." Results were also listed as mixed if the data indicated differing trends for the same factor at different categorical levels. Statistical significance of results was summarised from information provided in studies such as p-values or non-overlapping CIs; if any reported result was statistically significant, the overall result for that factor in that study was reported as being statistically significant. All associated factors with three or more synthesised measurement results were considered to have sufficient information to conclude the effect of associated factors with at least some certainty in the results. These results were summarised separately for TF and TT and included in an evidence map for ease of interpretation. Associated factors with only one or two reported results were considered to have insufficient information on which to draw any overall conclusions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Data synthesis measurement summary.

https://doi.org/10.1371/journal.pntd.0010272.t002

3.1 Overview of included studies

A total of 10,087 records were identified from the database search; of these, 6,211 were unique records. All 6,211 were screened, resulting in eligibility being assessed on 71 full-text records, of which 35 studies were retained (Fig 1). Of the 35 included studies [28–62], 29 included trachoma associated factors, of which eight also included spatial analysis methods [30,33,34,37,47,58,60,61]; six studies included spatial analysis methods only [32,35,46,49,54,56] (Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Study inclusion flowchart.

https://doi.org/10.1371/journal.pntd.0010272.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Summary of selected studies by survey type, study quality measures, trachoma outcome, and analysis type.

https://doi.org/10.1371/journal.pntd.0010272.t003

All but one study, which was published in a dissertation [60], were published in peer-reviewed journals. The publication period for included studies ranged from 1989 to 2020. Data collection for included studies was reported as early as 1986 [47,62], with the most recent data collection occurring in 2018 [59], with one study published in 1989 not reporting data collection dates [32]. Of the 35 studies, there were 33 distinct data collection activities. Three studies used the same data collected during the "Mali National Trachoma Survey" but had different objectives or outcomes and performed distinct data analysis activities with no overlapping variables [39,57,58]. The 35 studies included data from 26 countries. Most studies focused on a single country, while two involved a multi-country analysis [37,38]. The data used in the multi-country data analysis activities either did not include data used in the other 34 studies and/or had no overlapping variables with those other studies. TF prevalence ranged from 0% to 77% across 30 studies, while TT prevalence ranged from 0% to 23% across seven studies; one study did not report trachoma prevalence as a percent of the population. In most cases, the associated factor analysis did not disaggregate according to survey prevalence, treatment status, or past survey prevalence.

3.2 Study quality

Study quality results are summarised in Table 3. Most manuscripts reported data collected in a cross-sectional population-based survey, while two manuscripts reported a cohort study design [45,54]. The cross-sectional studies were chiefly population-based surveys with a two-stage sampling design for survey clusters (villages) and then households. In contrast, most of the studies in Brazil (five of six) [48–50,56,59] and one study in Burkina Faso were school-based surveys [45]. All but four studies were surveys solely focused on trachoma. Of the four multi-disease surveys, one had the trachoma component embedded as part of a Vitamin A assessment [42], two were part of a country adult blindness survey [41,61], and one was a joint trachoma and schistosomiasis survey [45]. Of the 25 population-based survey studies, 16 followed WHO survey guidance or similar; of these, five were conducted with support from the Global Trachoma Mapping Project (GTMP), which provided gold standard support for surveys conforming to WHO guidance from 2012–2016 [63]. The remaining nine population-based surveys measuring trachoma prevalence were stand-alone studies that did not specify the basis for their sampling strategy. The study population varied in the studies but was most commonly 1–9-year-olds for TF and ≥15 year-olds for TT. The majority of studies (32 of 35) reported specifically using the WHO simplified grading system for trachoma. One study did not specify the trachoma identification method used [49]. The use of 2.5× magnifying loupes was mentioned in 12 studies, with a further two using 3.5× or 4× loupes. All other studies, including the five GTMP studies, made no specific mention of the use of magnifying loupes, although it was standard practice for GTMP surveys. Overall quality for the 35 studies was evenly divided across the three categories: high (10), medium (13), and low (13).

3.3 Associated factors and trachoma studies

Twenty-nine studies included trachoma outcomes with associated factors (Table 4). The most common trachoma outcome studied was TF (25 studies) [28–31,34,38–43,45,47,48,50,51,53,55,59–62], while TT was analysed in seven studies [33,36,37,44,52,58,61,62]. Only four studies included analysis for both TF and TT outcomes [33,36,52,58]. One study in adults from China included "any trachomatous change" as their primary outcome; the one associated factor identified in this study (ruralness) was not included in the synthesised results [41]. Seven studies included only univariable analysis [29,44,47,48,51,55], eight only multivariable results [30,31,34,37–39,41,53], and 14 reported univariable and multivariable results [28,33,36,40,42,43,45,50,52,57–61]. Thus, the recording of results included 196 unique study, outcome, variable, and analysis type (univariable/multivariable) combinations.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Associated factor studies by trachoma outcome, analysis type, and variable category.

https://doi.org/10.1371/journal.pntd.0010272.t004

Overall, 59 different associated factors (S4 Text) were included in the 29 included studies. The infrastructure variable category had the most (23) associated factor variables across all the studies. Table 4 summarises the number of associated factors by variable category per study. The number of unique associated factors included in any individual study ranged from 1 to 25; ten studies included only one factor, while 17 had between two to six factors. The majority of studies (16) had at least one factor in the environment category.

Results are reported for each of the five variable categories, focusing on those associated factors included in three or more studies. Synthesised results for TF and TT all associated factors are included in S1 Table, while full results for each study are in S2 Table. Statistical significance of results is also presented as a binary outcome: "yes" if any of the reported results were significant and "no" if all results were not significant.

The source of covariate data included in the 29 studies with associated factors varied from geospatial raster datasets, survey-derived, government data or other survey data, meteorological station data, and direct global positioning system (GPS) reading. S4 Text summarises the sources of data for each variable. Survey-derived covariates include information collected as part of the same survey being analysed; this included many infrastructure variables such as water and sanitation coverage, presence of medical services, or paved roads. Other survey-derived variables included rural/urban classification. Government data sources or other survey data included information on school attendance and poverty measures; these data were linked to survey location by administrative unit information. GPS data collected during surveys were used for latitude, longitude, and in some cases, altitude measures.

3.3.1 Associated factor trends: Climate

The climate variable category included 14 unique variables used in studies examining TF. Only the specific variables for mean annual precipitation and mean annual temperature were used in three or more studies. The overall trend was that as mean annual precipitation increases, TF prevalence is lower; six of eight studies supported this conclusion (Fig 2) [30,33,34,36,58,60], of which five were statistically significant. The results for aridity index, relative humidity, air pressure, sunshine fraction, and monthly rainfall days were also in-line with this finding but had two or fewer studies using these variables (S1 Table). Mean annual temperature was included in five statistically significant analyses, with three indicating lower TF prevalence as temperature increases [30,39,45]. There were seven other temperature measures included across seven studies [28,33,36,39,45,58,60], but in all cases, two or fewer per specific variable (S1 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Associated factor trends for TF.

https://doi.org/10.1371/journal.pntd.0010272.g002

For studies looking at TT outcomes, aridity (lower aridity index means more arid area/drier area) and mean annual precipitation had enough results for synthesis. Data showed that areas with less aridity had lower TT prevalence (Fig 3) (nine of ten analyses, though only three of nine were statistically significant). Of note, all these analyses came from Flueckiger et al. 2019 [37], which analysed data from ten countries. Mean annual precipitation was included in four studies, but results were inconclusive [33,36,58,61]. All other TT results for climatic factors had two or fewer studies using the same variable, and thus no conclusions could be drawn.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Associated factor trends for TT.

https://doi.org/10.1371/journal.pntd.0010272.g003

3.3.2 Associated factor trends: Demographic

Within the demographic variable category, five variables were used across six studies examining TF. Areas considered to be "rural", as classified by the study survey design, had higher TF prevalence than "urban" areas (Fig 2) (four of five results from Brazil studies and three of five statistically significant) [47,48,50,59,60]. All other demographic variables did not have three or more studies that included the same measurement. The variable group that measured population density and size of the population in a village showed a trend of areas with decreasing population having higher TF prevalence [48,57,60].

Distance to nightlights, a proxy measure for population density or development, was used in a ten-country TT analysis [37]. Areas with fewer nightlights had higher TT prevalence (seven of ten, and three of seven were statistically significant) (Fig 3).

3.3.3 Associated factor trends: Environment

Within the environment category, five main variable groups were included in the TF studies. Altitude was by far the most common variable, being included in twelve analyses. Eight studies showed higher altitude areas had lower TF prevalence, and all associations were statistically significant (Fig 2) [28,29,31,33,40,43,55,58]. Ethiopia had eight of twelve results related to altitude [28–30,33,40,43,52,55], including six of eight in the "increasing altitude, lower TF prevalence" group. None of the other environment variables had three or more studies that included the same measurement.

For TT, six main variables were included in the studies. Altitude was included in four analyses [33,52,58,61] and showed mixed results with no clear trends (Fig 3). The sand/soil fraction, an indicator of dry or dusty earth, was equally split between higher or lower TT prevalence across ten countries [37].

3.3.4 Associated factor trends: Infrastructure

The infrastructure category had 23 unique variables. None was used in more than one analysis in studies with TF as the main outcome; however, several specific variables measured similar things with slight variation in the specific variable definition. Measures of accessibility or remoteness, including presence or distance to paved or any roads, as well as access to transportation or distance to a larger town, indicated that more remote areas or those with less access to roads had higher TF prevalence (Fig 2) (two of five associations were statistically significant) [30,51,60]. The variables for no medical care centre or pharmacy present in a village or increased distance from medical care were also associated with higher TF prevalence (two of six associations were statistically significant) [30,42,51,57]. On the other hand, the presence of a school, increased school density, smaller distance to school, or more school attendance in an area were associated with lower TF prevalence [57,60], though three of the four results came from the same study using different specific variables; all associations were statistically significant. Water or sanitation use, coverage, access, and availability were measured with nine different specific variables in nine studies. The overall trend indicates that less sanitation access (all associations were statistically significant) [38,53,60] and worse water access (two of three associations were statistically significant) [38,42,60,62] were associated with higher TF prevalence.

No infrastructure category variables were included in TT analyses.

3.3.5 Associated factor trends: Socioeconomic

The socioeconomic status category for studies with TF as main outcome included the fewest variables: six variables across two studies [57,60], and none met the threshold for inclusion in the synthesis of the results. See S2 Table for full results.

Smith et al. 2015 was the only study examining TT prevalence that included a socioeconomic associated factor [61].

3.4 Spatial analysis and trachoma studies

There were 14 studies identified in this systematic review that included spatial analysis [30,32–35,37,46,47,49,54,56,58,60,61] (Table 5). Of these, ten focused on TF only [30, 32–35,47,49,54,56,60], while two examined TT only [37,61], one looked at both TF and TT [58], and one included both a measure of TF and ocular Ct infection [46]. The level of analysis for the studies was predominately survey cluster (defined as village or community or school-based), while two studies performed household-level spatial analysis examining spatial trends within a survey cluster [32,54], and two included district-level analysis [30,33]. Most studies focused on a single country, while one TT study included ten countries [37]. As with the associated factor analysis, Ethiopia [30,33,37] and Brazil [47,49,56] were the most common countries included, with three studies each.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Spatial analysis studies by trachoma outcome, analysis level, and analysis type.

https://doi.org/10.1371/journal.pntd.0010272.t005

We grouped the spatial methods used in the studies into three main categories: exploratory spatial data analysis (ESDA), interpolation, and modelling. Half of studies (7 of 14) used spatial methods from at least two categories. The ESDA category was the most commonly used, employed in eleven studies [30,32–35,37,46,49,54,56,60], encompassing: Global Moran’s I, Local Moran’s I, Getis-Ord-Gi*, semivariogram, clustering analysis or correlograms, and Kulldorf spatial scan statistic (Table 5). The interpolation category includes three methods, each used once in three different studies [34,49,58]: kriging, kernel estimate, and unspecific spatial interpolation. Four analysis types were used across seven studies in the modelling category [35,37,46,47,54,60,61]; these included Bayesian, generalised linear model, mixed or random effect regression, and binomial logistic.