Evaluating scope and bias of population-level measles serosurveys: a systematic review and bias assessment

17 18

Measles seroprevalence data has potential to be a useful tool for understanding 19 transmission dynamics and for decision making efforts to strengthen immunization programs. In 20 this study, we conducted a systematic review and bias assessment of all primary data on 21 measles seroprevalence in low- and middle-income countries published from 1962 to 2021. 22 23

On March 9, 2022, we searched PubMed for all available data. We included studies 24 containing primary data on measles seroprevalence and excluded studies if they were clinical 25 trials or brief reports, from only health care workers, suspected measles cases, or only 26 vaccinated persons. We extracted all available information on measles seroprevalence, study 27 design, and seroassay protocol. We conducted a bias assessment based on multiple categories 28 and classified each study as having low, moderate, severe, or critical bias. This review was 29 registered with PROSPERO (CRD42022326075). 30 31 Findings: We identified 221 relevant studies across all World Health Organization regions, 32 decades and unique age ranges. The overall crude mean seroprevalence across all studies was 33 78.00% (SD: 19.29%) and median seroprevalence was 84.00% (IQR: 72.75 – 91.66%). We 34 classified 80 (36.2%) studies to have severe or critical overall bias. Studies from country-years 35 with lower measles vaccine coverage or higher measles incidence had higher overall bias. 36 37 Interpretation: While many studies have underlying bias, many studies provide data that can be 38 used to inform modelling efforts to examine measles dynamics and programmatic decisions to 39 reduce measles susceptibility. 40 41 Funding: Bill & Melinda Gates Foundation; Gavi, the Vaccine Alliance; US National Institutes of 42 Health 43 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint Research in Context 44 Evidence before this study 45 On August 20, 2023, we searched PubMed for systematic reviews published from January 1, 46 1980 to August 20, 2023 using the search terms “measles” AND “sero*”. We included studies if 47 they were a systematic review of measles seroprevalence data and excluded studies that did 48 not contain information on measles seroprevalence, were not systematic reviews, only included 49 data from persons within a subpopulation (e.g., pregnant people or healthcare workers), or were 50 of head-to-head laboratory comparisons of assay methodology. We identified one previous 51 systematic review, by Thompson and Odahowski, published in 2016 and including data through 52 mid-2014. That review identified 220 measles and/or rubella seroprevalence studies from all 53 countries globally. Study authors published a descriptive summary of seroprevalence trends by 54 age in a five select countries and a narrative summary of high-level epidemiologic trends in the 55 underlying data, including information available on maternal antibody waning. Beyond these 56 select summary findings, that study did not separately report seroprevalence from each study 57 identified in the analysis, nor did it include any information on study design or population-58 representativeness. While study authors noted general limitations related to the different 59

used across studies, they did not include any specific information on assay type, 60 selection biases or other characteristics that could influence the accuracy of results or include 61 data in a tabular format, which limits the utility of this study for subsequent analyses. 62 63 Added value of this study 64 Our study builds upon the known body of data on measles seroprevalence from low- and 65 middle-income countries in multiple ways. First, we included data published up to December 31, 66 2021 and from non-English language studies. Second, we extracted all available relevant 67 information on study design characteristics and assay protocol used in each study to measure 68 seroprevalence. Then, we constructed a bias assessment framework and conducted a bias 69 assessment across multiple categories (study selection of participants, measurement tool and 70 classification of immunity, and reporting of results) to classify the underlying bias in each study. 71 Finally, we compared seroprevalence estimates across regions and bias levels, and bias levels 72 among various study location characteristics. 73 74 Implications of all the available evidence 75 Accounting for study design and seroassay protocol used in serosurveys can influence 76 interpretation of population-level seroprevalence estimates. Our systematic review and bias 77 assessment provides an updated landscape of serological studies and highlights key biases in 78 the current literature. It provides a repository of measles seroprevalence data, along with 79 corresponding critical information on factors that influence population-representativeness and 80 overall sensitivity of the measurement assay used in each study, that can be used to inform 81 measles susceptibility estimates useful for planning targeted vaccination efforts. 82 83 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint

84 Measles remains a substantial cause of global morbidity and mortality1, especially in low- and 85 middle-income settings where over 99% of measles cases and deaths occur2, despite the 86 availability of a safe and effective vaccine3. Because ongoing measles transmission can be 87 maintained if herd immunity (i.e., when the proportion of the population immune is sufficient to 88 limit disease spread) has not been reached and sustained, estimating the proportion of people 89 susceptible within a community is essential to plan immunization programs and assess future 90 risk of measles outbreaks and deaths. However, due to factors such as timeliness of and age at 91 vaccination4, disruptions to cold chains5, a lack of seroconversion in specific subpopulations 92 (e.g., among persons living with human immunodeficiency virus (HIV)6), and variable 93 surveillance systems across locations and time, inferring population-level measles immunity 94 from a combination of vaccination coverage and case notifications can be challenging7. 95 Alternatively, serosurveys can provide a snapshot of immunity gaps that remain in a community 96 by determining population-level prevalence of IgG antibody levels above specific thresholds that 97 suggest clinical protection against disease. 98 99 As such, seroprevalence data can be used as tools to guide decisions to and strengthen 100 immunization programs, as inputs to dynamic models of disease transmission, and additionally 101 to provide insights into vaccine field effectiveness and assessment of case ascertainment 102 rates7,8. The interpretation of seroprevalence data is complicated, however, because of the 103 potential for bias. Some of this bias can be due to inadequate sensitivity of laboratory assays9 104 and/or specimen types10 used for measuring antibody levels. Additionally, bias from assay 105 procedures can be suspected when protocols or commercial details are not reported or if no 106 quality control was performed. Furthermore, population-based surveys have the potential for 107 additional bias to be introduced in the selection of participants or from lack of 108 representativeness of the selected sample from the community. 109 110 Beyond understanding the selection processes and laboratory assays used, it is critical to also 111 consider how results of the serosurveys are reported. Considerations include what threshold of 112 antibody titer was used as a correlate of clinical protection and how some tests report 113 indeterminate results. In order to responsibly use and accurately interpret seroprevalence data 114 for decision making or for modelling exercises, these issues need to be transparently 115 acknowledged and discussed. 116 117 A more in-depth understanding of available seroprevalence data across locations and time, as 118 well as the related implications, is critical for using these historic data to calibrate models used 119 to inform decision making for immunization program strengthening, especially in low- and 120 middle-income countries (LMICs) that face the highest ongoing measles burden. To fill these 121 gaps, we first conducted a systematic review of literature reporting measles seroprevalence 122 data published through 2021 and extracted information on key study and assay information. 123 Then, we developed a pilot bias assessment tool to assess the risk of bias in each study across 124 the following categories: study selection of participants, measurement tool and classification of 125 immunity, and results reporting. 126 127

128 Search strategy and selection criteria 129 This study follows PRISMA guidelines (Supplementary Tables 1-2) and was registered with 130 PROSPERO (CRD42022326075). We performed a systematic review of published literature in 131 any language containing information on population-level measles seroprevalence in LMICs. We 132 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint searched PubMed on March 9, 2022 for primary data published through December 31, 2021 133 using the following search string: 134 135 (((Measles) AND (seroprevalence OR sero-prevalence OR seropositive OR sero-positive 136 OR seronegative OR sero-negative OR seroepidemiology OR sero-epidemiology OR 137 seroprofile OR seroimmunity OR sero-immunity)) 138 OR ("Measles/epidemiology"[MeSH] AND (antibod* OR serolog*))) 139 AND ("1900"[Date - Publication] : "2021"[Date - Publication]) 140 141 One individual (ANS) screened titles and abstracts for each study in the search results. For 142 relevant studies, one of multiple individuals (ANS, HF, IP) reviewed the full-text of each to 143 determine their inclusion or exclusion. We included studies that contained original data on 144 measles antibody prevalence and excluded studies if they only contained data from high-income 145 locations (as based on WorldBank 2021 income classifications11), did not contain data on 146 measles IgG antibody, were based on non-original data or from non-human subjects, contained 147 only results from laboratory assay development or clinical trials (including studies only 148 containing information on vaccinated persons), studied a target population of only health-care 149 workers or active measles cases, or were a review, abstract, letter, editorial or brief report. 150 151 Following full text review, for each study that met our inclusion and exclusion criteria, we 152 extracted the following data: study setting, study design and type (including information on 153 planned, achieved (i.e., how many persons were reached via sampling), and reported (i.e., how 154 many persons were represented in final study metrics) sample sizes), population demographics 155 (including income and representativeness), type of specimen collected, serologic assay details 156 (including type, name, and inclusion of a reference preparation), antibody threshold used for 157 seropositivity and/or seroprotection (if relevant), and measures of proportion seropositive, 158 seronegative, or indeterminate with accompanying uncertainty. We extracted data into a 159 Microsoft Excel workbook and for seroprevalence measure, we recorded the most granular 160 levels for relevant strata (i.e., by age, vaccination status, infection history, etc.) presented in 161 each study. 162 163 Bias assessment 164 Following extraction of all available data, we developed a comprehensive bias assessment tool 165 and applied the tool to characterize the level of bias across each study. Our tool, modified from 166 the ROBINS-I tool12, considers bias across the following categories, with associated indicators: 167 study selection of participants, measurement tool and classification of immunity, and reporting of 168

(Supplementary Figures 1-3). We classified the level of bias across each category to be 169 either low, moderate, severe, or critical. We then finally assessed the overall level of bias as 170 low, moderate, severe, or critical for each study by taking the mean score of the category-171 specific classifications. 172 173 To assess bias among study selection of participants, we considered whether the study design 174 used a random process for sample selection, if a study relied on a convenience sample, was 175 restricted only to a subset of the population (e.g., only included pregnant women or cancer 176 survivors), and reporting of planned, achieved, specimen, and final sample sizes. To assess the 177 level of bias among the measurement tool and classification of immunity, we considered 178 whether assay protocol, name, or references were provided, if internal or external validation or 179 quality control was performed, and if there were other known factors known to decrease 180 sensitivity or specificity. These factors included using oral fluid as specimens13, using a 181 hemagglutination inhibition (HI/HAI) assay13, or using the Whittaker enzyme-linked 182 immunosorbent assay (ELISA)14. Last, for bias among reporting of results, we considered 183 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint whether a known threshold was used for determining protective titer levels, including metrics of 184 uncertainty with seroprevalence estimates, and, if an enzyme immunoassay (EIA) or ELISA was 185 used, whether and how equivocal results were handled and reported. 186 187 We characterized the overall level of bias in each study using the following criteria. For each 188 category of bias studies were given a numeric score: low bias was assigned a score of 1, 189 moderate a score of 2, severe a score of 3, and critical a score of 4. We took the mean of 190 scores across all three categories. Studies with a mean score below 1.5 were characterized to 191 have low overall bias, between 1.5 and 2.5 to have moderate overall bias, between 2.5 and 3 to 192 have severe bias, and more than three to have critical bias. 193 194 We converted all metrics reported to proportion seropositive and then used R version 5.4.0 to 195 compute summary metrics and make figures. For studies reporting seropositive and 196 indeterminate/equivocal results independently, we did not include indeterminate results in the 197 numerator of our overall seroprevalence calculation. We compared data availability by decade 198 and bias level. We additionally investigated bias levels across time and region and assessed 199 bias levels across locations with higher and lower first-dose measles-containing vaccine (MCV1) 200 coverage15 and higher and lower estimated annual measles incidence16 in the year from which 201 study data was collected. 202 203 Role of the funding source 204 The Bill & Melinda Gates Foundation, Gavi, the Vaccine Alliance and the US National Institutes 205 of Health had no role in study design, data collection, data analysis, data interpretation, or the 206 writing of the report. All authors had access to the data and the corresponding author had final 207 responsibility to decide to submit for publication. 208 209

210 Systematic review 211 From our search, we identified 2032 studies for screening (Figure 1). Following screening, we 212 excluded 1116 studies that did not meet our search criteria. For the remaining 916 studies, we 213 assessed the full-text articles for inclusion. We identified 221 studies for inclusion and extracted 214 information on measles seroprevalence, study design, and seroassay (link to zenodo file once 215 uploaded). Studies were published between 1962 to 2021, including seroprevalence surveys 216 conducted between 1953 and 2019. 217 218 Among 182,789 persons sampled across all studies, age groups, and years, the crude mean 219 measles seroprevalence was 78.00% (SD: 19.29%) and median seroprevalence was 84.00% 220 (IQR: 72.75 – 91.66%). 221 Across regions of the World Health Organization (WHO), there were 43 studies containing data 222 from the African Region, 47 from the Eastern Mediterranean Region, 35 from the European 223 Region, 25 from the Region of the Americas, 20 from the South-East Asia Region, and 73 from 224 the Western Pacific Region (Figure 2). There were 24 studies that represented data collected 225 before 1980, 32 studies from 1980 to 1989, 29 studies from 1990 to 1999, 55 studies from 2000 226 to 2009, and 83 studies from 2010 to 2019. 178 studies (80.5%) contained age stratified results 227 across 531 unique age ranges. 228 229 Bias assessment 230 Table 1 shows results of our bias assessment for each included study. For overall bias, we 231 classified bias as low in 12 (5.4%) studies moderate in 129 (58.3%), severe in 58 (26.2%) and 232 critical in 22 (10.0%). No studies had low or critical bias across all the categories of study 233 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint selection of participants, measurement tool and classification of immunity, and reporting of 234

(Table 1). 235 236 For study selection of participants, we identified 15 studies with low bias, 181 with moderate 237 bias, 23 with severe bias, and 2 with critical bias. 81 studies used a random sample selection 238 method. 117 studies with convenience samples used a restricted, non-representative sample 239 (i.e., only among a specific subgroup of the population, such as persons living with HIV). 2 240 studies did not report the final sample size, and of the 81 samples that used a random sample 241 selection method, 45 reported the planned sample size, and 15 additionally reported the 242 planned, achieved, and specimen sample sizes. 243 244 In measurement assay and classification of immunity, we identified 19 studies with low bias, 130 245 with moderate bias, 46 with severe bias, and 26 with critical bias. Across the three categories of 246 bias assessment, measurement assay and classification of immunity had the highest number of 247 studies classified as having critical bias, largely due to absence of information on assay protocol 248 details, commercial kit name or other appropriate citation describing the underlying methods. 249 195 studies provided details on the assay protocol or commercial kit name, and 25 studies 250 conducted internal or external validation or quality control. 6 studies specified that samples were 251 oral fluid specimens and 30 studies specified that samples collected were dried blood spots. 252 253 54 studies used an HI/HAI assay, 139 used an EIA or ELISA, 13 used a plaque reduction 254 neutralization test (PRNT), 6 used a multiplex bead assay, and 11 used other or undescribed 255 assay types. We noted changing temporal trends of types of seroassays used. While EIA, 256 ELISA and PRNT assays were used in even distribution across all studies examined, there was 257 no study published after 2001 that utilized an HI/HAI assay, and all studies using a multiplex 258 immunofluorescent assay were conducted in 2013 or later. 259 260 We identified 20 studies with low bias, 63 with moderate bias, 70 with severe bias, and 18 with 261 critical bias in reporting of results. 155 studies reported a threshold to define seroprevalence. 262 Among the 139 studies that used an EIA or ELISA, 30 studies reported equivocal results 263 separately or included with seropositivity results and 1 study excluded equivocal results and 264 they were less than 5% of the overall sample. Finally, 59 studies reported metrics of 265 seropositivity or seronegativity with any accompanying uncertainty. 266 267 Seroprevalence trends 268 The crude median seroprevalence estimates from studies in the Western Pacific Region was 269 88.3% (IQR: 79.2 – 93.4%), in the Eastern Mediterranean Region was 87.2% (IQR: 81.3 – 270 93.2%), the European Region was 82.0% (IQR: 77.8 – 89.0%), in the Region of the Americas 271 was 78.4% (IQR: 60.7 – 93.0%), in the African Region was 77.6% (IQR: 60.7 – 89.9%), and in 272 the South-East Asia Region was 66.8% (IQR: 47.4 – 88.4%). Trends in seroprevalence and bias 273 vary by decade (Figure 3). The median seroprevalence was lower in studies from 2010 to 2019 274 than those conducted before 1980 (i.e., the pre-vaccination era). Crude seroprevalence from 275 studies conducted before 1980 was 90.5% (IQR: 67.8 – 93.3%), from 1980 to 1989 was 78.6% 276 (IQR: 57.8 – 90.7%), from 1990 to 1999 was 88.3% (IQR: 60.7 – 92.6%), from 2000 to 2010 277 was 80.4% (IQR: 65.6 – 88.2%), and from 2010 to 2019 was 84.6% (IQR: 78.3 – 92.9%). 278 Among 31 country-years with studies containing critical bias, 23 (74%) occurred in earlier time 279 periods (i.e., before 1980 and between 1980 and 1989). In the 159 country-years with studies 280 containing low or moderate bias, 96 (60%) have occurred between 2010 and 2019. 281 282 We additionally compared the overall bias levels for each country-year of the studies to the 283 MCV1 coverage and measles incidence from the same country-year (Figure 4). Generally, 284 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint studies in countries and years in 1980 or later with lower MCV1 coverage and higher measles 285 incidence had more bias compared to studies from countries and years with higher MCV1 286 coverage and lower measles incidence (p < 0.001, in proportional odds logistic regression 287 models for both MCV1 coverage and incidence). Among 109 studies from countries and years 288 with MCV1 coverage greater than 80%, 93 (85%) had low or moderate overall bias, and from 289 the 58 studies from countries and years with MCV1 coverage of 80% or lower, 34 (58%) had 290 low or moderate overall bias. A similar trend persisted across studies in countries and years 291 with high incidence – 103 of 122 (84%) studies in countries and years with average annual 292 reported measles incidence less than 5 per 1000 persons had low or moderate overall bias, and 293 24 of 49 (49%) of studies in countries with annual measles incidence of 5 per 1000 persons or 294 greater had low or moderate overall bias. 295 296

297 To identify the scope of measles seroprevalence data, we conducted an updated systematic 298 review of serosurveys to identify primary data sources and characterized underlying bias across 299 these studies. The resulting data repository from our investigation along with information on 300 factors related to underlying bias per study could contribute to analyses of measles dynamics 301 among low- and middle-income countries. We identified serosurveys available in each decade, 302 WHO region, and across a wide variety of ages, which could be useful when modelling location-303 , time-, and age-specific estimates of measles transmission and susceptibility. Despite this 304 variation, there were locations for which very few or no serosurveys have been conducted – 305 mainly in the African Region – which contribute to knowledge and data gaps to inform high-306 quality modelling and analyses. 307 308 Additionally, our study provides insight to issues to consider when designing and reporting a 309 seroprevalence study to ensure that the highest quality surveys are conducted and that 310 complete, accurate and transparent reports are generated. The number of available measles 311 seroprevalence studies has increased in the last few decades compared to periods before the 312

of national measles vaccination programmes in LMICS. This trend provides the 313 opportunity for researchers to examine the impact of vaccination programs on ongoing 314 susceptibility within the population represented in each study. However, we found that locations 315 with high annual measles incidence and lower MCV1 coverage tend to have not only less 316 studies conducted, but also higher bias – this is understandable given that coverage tends to be 317 lower in the most difficult settings such as remote and/or conflict-affected regions, where 318 surveys are especially challenging to conduct. Research and programmatic teams planning 319 seroprevalence studies, especially among persons living in these vulnerable communities, could 320 use the framework presented in this study as a starting point to determining the feasibility and 321 cost of conducting a high-quality seroprevalence survey and consider alternative ways to invest 322 the funds (e.g., in strengthening ongoing surveillance of coverage and disease incidence). 323 324 More recently, there have been examples of high-quality serosurveys, such as a nationally 325 representative survey in Zambia17, that have been conducted and used for informative 326 modelling. Given the complexity, time, and expense of these surveys, it is worthwhile to make 327 the most of high-quality surveys that are being conducted for different infections and funded 328 through a variety of different programs. This serosurvey in Zambia, for example, leveraged 329 residual sera from the Zambia Population-Based HIV Impact Assessment (ZAMPHIA) study18 330 originally collected to estimate HIV incidence and viral load. Applications of such data extend to 331 innovative modelling efforts to estimate subnational and age-specific seroprevalence estimates 332 as well as national level outbreak risk17. That study serves as an example of the potential to 333 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint leverage other major population surveys and to use high quality seroprevalence estimates to 334 inform evidence for decision making. 335 336 More studies had low or moderate bias compared to severe or critical bias among the 337 categories of selection of study participants and measurement tool and classification of 338 immunity. For the category of reporting of results, more studies had severe or critical bias levels 339 than low or moderate bias levels. Overall, we found that less than 10% of studies had low 340 overall bias, suggesting that the quality of conduct and reporting of seroprevalence studies has 341 substantial potential for improvement. 342 343 While interpreting seroprevalence estimates identified by our review, it is essential to also 344 consider the associated sensitivity and specificity of the seroassays used in studies along with 345 the route of induced immunity (i.e., from vaccination or natural infection). For example, HI/HAI 346 assays are often less sensitive than other types13. If HI/HAI assays are used in a population with 347 mainly vaccine-induced immunity, seroprevalence results may be underestimated. However, 348 since HI/HAI assays were historically used more frequently, during an era with less vaccine-349 derived immunity and subsequently higher natural immunity affording higher antibody levels, 350 assay sensitivity might not be as important to consider. In our bias assessment in the category 351 of measurement tool and classification of immunity, we defined factors that influence assay 352 specificity and sensitivity as either (1) using an HI/HAI assay, (2) using the Whittaker 353 commercial ELISA kit, or (3) using oral fluid samples. However, the utility of this specific 354 contribution to our bias assessment might be subject to the specific study setting, vaccination 355 program implementation and success, and underlying measles epidemiology. 356 357 Our study has several limitations. First, we were unable to fully synthesize results of our 358 systematic review in a meta-analysis or other stratified analysis by age, location, or year. This 359 was due to the differing study populations, regions, time periods, and age groups presented in 360 studies identified in this review as well as the varying degrees of bias characterized to be 361 present across studies. These results can serve as the basis for future models that synthesize 362 the data while also accounting for underlying measles infection dynamics, vaccination coverage 363 and population structures for each individual study setting, which was out of the scope of our 364 analysis. 365 366 Secondly, we were constrained by the information reported in each publication. Without 367 adequate reporting, we assumed the highest level of associated bias whenever appropriate. For 368 example, if a study did not specifically note if they used an international reference preparation, 369 we assumed they did not use one. This may have led us to classify studies as having higher 370 bias in relevant categories than might have been the case if all available information had been 371 included in the publication – it possible that some details were omitted to meet restrictions on 372 word counts, for example. As such, there might be great utility in the widespread use of 373 standardized reporting expectations for ongoing and future seroprevalence studies. 374 375 Next, we did not consider sample size in our assessment of bias. Since the impact of sample 376 size on the reliability of point estimates from seroprevalence studies should be reflected in the 377 provided uncertainty interval, we considered the inclusion of such in our bias assessment. We 378 did not however further assess the implications of smaller or wide interval spans of if they were 379 presented or whether point estimates or uncertainty intervals were adjusted or standardized for 380 population demographics or other factors. Finally, there are likely additional sources of bias that 381 are more difficult to ascertain objectively, such as potential issues with specimen storage and 382 laboratory capacity, practices, and quality. 383 384 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint Our study strengthens the understanding of the availability and bias among measles 385 seroprevalence studies in low- and middle-income countries by identifying primary sources of 386 measles seroprevalence studies and conducting a bias assessment of the associated data. Our 387 framework for assessing bias could provide a foundation for further work by relevant agencies 388 and interested partners to develop a tool for use in planning and reporting future surveys. This 389 work can be a vital tool to be used during modelling exercises, planning immunization-based 390 interventions, and ultimately, to make informed decisions to reduce preventable measles 391 morbidity and mortality. 392 393 Contributions 394 ANS, MJ and JFM conceived and planned this study. ANS, FC, DR, MJ and JFM designed the 395 bias assessment framework. ANS, HF and IP screened and extracted studies. ANS made 396 tables and figures. ANS wrote the first draft of the manuscript and all authors contributed to 397 subsequent revisions. 398 399

400 401 1 Minta, A. A . et al . Progress Toward Regio nal Measles Elimina tion - Worldwid e, 20 00-2021. 402 MMWR Mor b Mor tal Wkly R ep 71, 1489-1495 (2022). 403 https://doi. org:10.15585/mmwr.mm7147a1 404 2 Global bu rden of 369 diseases and injuri es in 204 countries and ter rito ries, 1990- 2019: a 405 systematic analysis for the Global B urde n of Disease Study 2019. Lance t 396 , 1204-1222 (2020 ). 406 https://doi. org:10.1016/s0140-6736(20) 30925-9 407 3 Keja, K., Chan, C., Hayden, G. & Hend ers on, R. H. Ex panded p rogramme on immu nization . Worl d 408 Healt h Stat Q 41, 59-63 (1988). 409 4 Hughes, S. L. et al. The effect of time sinc e measles vaccination and age a t first do se on measles 410 vaccine effectiveness - A systematic revie w. Vaccine 38, 460-469 (2020). 411 https://doi. org:10.1016/j.vaccine.2019 .1 0.090 412 5 Kumru, O. S. e t al. Vaccine ins tabili ty in the cold chain: mechanisms, an alysis and formulation 413 strat egies. Bi olo gicals 42, 237-259 (2014) . https ://doi.org:10 .1016/j.biologicals.20 14.05.007 414 6 Mutsae rts, E. e t al. S afety and Immunog enicity of Measles Vaccinati on in HIV-Infected and H IV-415 Exposed Uninfec ted Childre n: A Systema tic Review and Me ta-Analysis. EClini calM edicin e 1 , 28-416 42 (2018). https://doi.org:10.1016/j.eclin m.2018.06.002 417 7 Winte r, A. K. e t al . Ben efits and Challeng es in Using Seropr evalence Dat a to I nfor m Models for 418 Measles and R ubella Elimina tion. J Infe ct Dis 218 , 355-364 (2018 ). 419 https://doi. org:10.1093/infdis/jiy137 420 8 Cutts, F. T. & Hanson, M . Ser oepid emiolo gy: an underused tool for designing and monitoring 421 vaccination progr ammes in low- and middle-income countri es. Tro p Med Int He alt h 21, 1086-422 1098 (2016). https://doi.org:10.1111/tmi .12737 423 9 Lutz, C. S. et al . Comparison of measl es I gG enzyme immunoassays (EIA)yi&Cversus plaque 424 reductio n neut raliza tion test (PRNT)yi&Cfor measuringyi&Cmeaslesyi&Cs erost atus : a syste matic review of 425 head-to-he ad analyses of measles IgG EI A and PRNT. BMC Infect Dis 23, 367 (2023). 426 https://doi. org:10.1186/s12879-023-08199-8 427 10 Hutse, V. e t al. O ral fluid for th e serologi cal and molecular diagn osis of measles. Int J Infec t Dis 428 14, e991-997 (2010). https://doi.org:10.1 016/j.ijid.2010.06.009 429 11 World Bank . The Worl d by Inc ome a nd R egion , (2021). 431 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint 12 Stern e, J . A. e t al. R OB INS-I: a tool for ass essing risk of bias in non-randomised stu dies of 432 interven tions . BMJ 355 , i4919 (2016). htt ps://doi.org:10.1136/bmj.i4919 433 13 World He alth O rganiza tion . Man ual for t he Labor ator y-base d Surveilla nce of Mea sles, Rubella , 434 and Co nge nit al Ru bella Syn drom e , ( 436 14 Cutts, F. T. e t al. Immunog enicity of high-tite r Edmonston-Zagreb me asles vaccine in human 437 immunodeficiency virus-infected childre n in Kinshasa, Zaire. J Infec t Dis 167 , 1418-1421 (1993 ). 438 https://doi. org:10.1093/infdis/167.6.141 8 439 15 WHO. WHO/UNICEF esti ma tes of nati on al immu niza tio n cov erag e , 440 < htt ps://www.who.int/teams/immunizat ion-vaccines-and-biologicals/immunizati on-analysis-441 and-insights/global-monitoring/immuniz ation-coverag e/who-unicef-estimates-of -national-442 immunization-coverag e > (2022). 443 16 Eilertson , K. E., Fricks, J . & Ferrari , M. J. E stimation an d predic tion for a mechanis t ic model of 444 measles tr ansmission using particle filt eri ng and maximum likelihood es timation . Stat M ed 38, 445 4146-4158 (2019 ). https://doi.org:10 .10 02/sim.8290 446 17 Carcelen, A . C. et al . Leveraging a na tion al bioreposi tory in Zambia to ass ess mea sles and rubell a 447 immunity gaps across age and space. Sci Rep 12, 10217 (2022). https://doi.org:10 . 1038/s41598-448 022-14493-3 449 18 Ministry of Healt h Zambia. Zambia Popul ation-base d HIV Impact Ass essment (ZAMPHIA) 2016: 450 Final Repo rt. (Lusaka, Minis try of Health , Februa ry 2019). 451 452 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint Tables 453 454 Table 1. Overall and categorical bias classifications. 455

of bias assessment in each of three categories (study selection of participants, 456 measurement tool and classification of immunity, and reporting of results), and mean level of 457 bias per study. 458 459 Level of bias Mean Study selection of participants Measurement tool and classification of immunity Reporting of

Low 12 15 19 20 Moderate 129 181 130 63 Severe 58 23 46 70 Critical 22 2 26 18 460 461 462 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint Figures 463 464 Figure 1. PRISMA diagram. 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 Records identified from: PubMed (n = 2032) Records removed before screening: Duplicate records removed (n = 0) Records marked as ineligible by automation tools (n = 0) Records removed for other reasons (n = 0) Records screened (n = 2032) Records excluded (n = 1116) Reports sought for retrieval (n = 916) Reports not retrieved (n = 0) Reports assessed for eligibility (n = 916) Reports excluded: High-income country (n = 447) No measles data or serology data (n = 26) Research, vaccine efficacy study, trial (n = 92) Literature review (n = 7) Non-original data (n = 28) Outbreak or case-testing (n = 53) Abstract, editorial, brief report (n = 32) Only included health-care workers (n = 10) Studies included in review (n = 221) Identification of studies via databases and registers Id en tifi ca tio n Sc re en in g In cl ud ed . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint Figure 2. Number of serosurveys with data included per country. 505 Map of number of studies per country with available data identified by systematic review. 506 507 508 509 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint Figure 3. Measles seroprevalence by time period and overall bias level. 510 Beeswarm plot of measles seroprevalence by time period. Each point represents one country-511 year of data per study and are colored by overall bias level. Black lines represent the median 512 observation across each decade. 513 514 515 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint Figure 4. Overall bias level by MCV1 coverage and annual measles incidence. 516 Each point represents each country-year represented across all studies, colored by overall bias 517 level, by MCV1 coverage and annual estimated measles incidence. 518 519 520 . CC-BY 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted August 29, 2023. ; https://doi.org/10.1101/2023.08.29.23294789doi: medRxiv preprint