Serological Surveys for complementing assessments of vaccination coverage in sub-Saharan Africa : A systematic review

Serosurveys of biomarkers of infection/vaccination are widely Background: used for evaluating vaccine-induced immunity and monitoring the effectiveness of immunisation programmes in developed countries. In sub-Saharan Africa (sSA) where vaccination coverage (VC) estimates are often incomplete, inaccurate and overestimate effective population immunity, the use of serosurveys is limited. We conducted a review of the use of serosurveys to Methods: assess/complement assessments of VC in sSA by searching electronic databases (PubMed, Embase, Web of Science, Popline, Ovid and Africa Wide Information) for English language articles published from 1  January 1940 to 31  January 2017. We also searched the references of retrieved articles. SSA was defined as all of Africa excluding the countries in North Africa. We included only articles that measured VC and assessed the quality of these studies using the Newcastle-Ottawa Scale. We found 1056 unique records, reviewed 20 eligible studies of which Results: just 12 met our inclusion criteria. These 12 studies were serosurveys of measles, tetanus, polio and yellow fever. Antibodies induced by natural infection confounded serological test results and there was significant discordance between vaccination history and the presence of antibodies in all except for tetanus vaccine. No study looked at Hepatitis B. Serosurveys for tetanus or tetanus containing vaccines may be Conclusions: directly useful for ascertainment of vaccination exposure or reliably complement current survey methods that measure VC. Given the limited experience in using serosurveys for this purpose in sSA, well-designed serosurveys of tetanus and possibly hepatitis B are required to further validate/evaluate their performance.


Introduction
Serological surveys (serosurveys) provide invaluable insight into the natural history and epidemiology of infection and can be used to: assess the effectiveness of vaccine campaigns, determine proximity to theoretical thresholds for disease elimination, estimate the burden of disease for immunising and/or chronic infections, and identify gaps in population immunity to inform interventions such as supplementary immunisation activities [1][2][3][4] . As a tool in seroepidemiology, serosurveys have been a significant component of disease control and elimination strategies for acute viral vaccine preventable disease (VPDs) like measles, rubella, and polio in high-income countries. They have also been used to demonstrate the continued effectiveness of vaccination programmes 5 , and to guide revisions/adjustments to vaccination strategies for pertussis, diphtheria and Haemophillus influenza B (Hib) [6][7][8] .
Despite some of the well-known limitations of serological surveys, it is believed they can play a role in improving the accuracy of estimates of vaccination coverage (VC), or be used independently to assess VC particularly in developing countries 9 . Receipt of the third dose of the combined diphtheria, tetanus and pertussis vaccine (DPT3) assessed at 1 year of age is currently the most widely used proxy for childhood vaccination coverage worldwide 10 . VC is also an indicator of health care delivery, a performance metric for national immunisation programmes, and a global health target indicator. It is typically obtained from administrative records (health facility registers and vaccine doses consumed) or household surveys of retained immunisation cards and/or caregiver recall of childhood vaccination history. These methods have well known limitations that frequently result in over-or underestimates of VC 9 . Since receipt of a vaccine does not always induce protective immunity, these methods report a crude VC, and there is frequently a gap between crude (beyond those related to the intrinsic properties of the vaccine) and effective VC given the recurrent outbreaks of VPDs in settings reporting high crude VC 11,12 .
Ideally, overall evaluations of national immunisation programmes in low and middle income countries (LMICs) and for effectiveness of specific vaccination programmes should include direct measurement of population immunity in serological surveys. Compared to high-income countries, there appears to be very limited experience with the use of serological surveys in sub-Saharan Africa (sSA) LMICs to complement measures of crude VC and determine effective VC or population immunity. Consequently, there is very little application of seroepidemiology in monitoring the effectiveness of vaccination programmes and for informing vaccine policy in sSA.
This overall aim of this study is to assess the extent to which serological surveys have been carried out in sSA to complement assessments crude VC. We carried out a systematic review of studies reporting on vaccinated populations who also had samples taken for serological tests with a view of comparing the test results to records of vaccination in sSA.

Literature search and selection criteria
This review was conducted in accordance with PRISMA guidelines (see checklist, Supplementary File 1) 13  OR "coverage"[All Fields]))) OR (("vaccination"[MeSH Terms] OR "vaccination"[All Fields]) AND ("AHIP Cover"[Journal] OR "coverage"[All Fields]))) AND ("Africa south of the Sahara"[MeSH Terms] OR ("Africa"[All Fields] AND "south"[All Fields] AND "Sahara"[All Fields]) OR "Africa south of the Sahara"[All Fields] OR ("sub"[All Fields] AND "Saharan"[All Fields] AND "Africa"[All Fields]) OR "sub Saharan Africa"[All Fields]). The same query was used to search the EMBASE database.
We included English language articles reporting primary data from a serological survey that included assessments of VC from research conducted in sSA. We defined sSA as the whole of Africa except Northern Africa. Reviews, reports of serosurveys or vaccination coverage alone were not included. Full text copies of the relevant articles meeting our inclusion criteria were obtained. In addition, their references were individually checked to identify other related papers meeting the study inclusion criteria (see Figure 1). Retrieved citations were extracted to Endnote™ X7 (Thomas Reuters, PA, USA). The following variables were extracted from each included article-country/setting of study, year of study, sample size, age of participants, vaccine administered, biological specimen used for serology, type of serological test done, numbers who received vaccines from vaccine records and numbers of positive serology results. The key outcome of interest was vaccination coverage assessed by receipt of vaccines compared to results of serological tests.

Quality assessment
As recommended by the Cochrane Public Health Review Group 14 and used in other reviews 15 , the Newcastle Ottawa Scale (NOS) was used to assess the quality of each eligible study, to ascertain the risk of bias; and internal and external validity 16 . Studies were assessed in three domains: selection of the study groups, comparability of the study groups, and the ascertainment of the outcome of interest and a point system is awarded for each item evaluated 16 . With a total of 10 points, studies with ≥8, 6-7, 4-5, and ≤3 points respectively were evaluated as very good, good, satisfactory and unsatisfactory studies. (see Supplementary File 2).

Data synthesis
Since the studies eligible for inclusion in this review were few, and showed significant heterogeneity in study design and ascertainment of outcomes, a meta-analysis for derivation of pooled estimate of effect was not possible. Instead, a narrative approach was used to describe findings from each study grouped by outcome measures.

Search results
As shown in Figure 1, literature search yielded 1056 articles but only 12 of them were relevant to this review. These 12 were from Western (Nigeria 17,18 , Gambia 19 , Mali 20 and Ivory Coast 21 ); Eastern (Ethiopia 22,23 , Kenya 24 , Tanzania 25 , and Sudan 26 ) and Southern (Malawi 27,28 ) regions of Africa. There were no studies from the Central African region. Although there were publications from all but one region of sSA, only 9 of 39 countries were represented.
The reviewed articles reported serosurveys of 3 viral replicating vaccines -polio, measles, yellow fever and one toxoidtetanus (see Table 1). There were a total of 10 studies on measles 17-19,21-25,28,29 , 1 on polio 19 , 1 on yellow fever 26 and 3 on tetanus 19,20,22 . In these surveys, results were compared to vaccination history ascertained by caregiver recall, presence of a known vaccination scar and/or vaccination record (card or clinic records). Table 2 summarizes the overall and category-specific quality of the included studies, along with some limitations. A rating of very good was given to three studies 21,22,24 , majority i.e. 8 studies 17,19,20,23,[25][26][27][28] were assessed as good while one 18 was rated just satisfactory.

Measles vaccine
In the study by Breman et al. 21 in Cote d'Ivoire, pre-campaign vaccination history and crude coverage were determined by history of vaccination and/or the presence of a scar following smallpox vaccination. The pre-campaign crude VC was 53.6%, but serological surveys showed 98.3% (243/247) and 91.7% (297/324) of children 6-8 and 9-24 months old respectively lacked measles antibodies. About 2.5% of children who reported not having received any vaccination had evidence of a vaccination scar. Among the children whose families reported a history of measles or vaccination before the recent campaign, only 45.5% (27/62) had measles antibodies. For susceptible children vaccinated during the campaign, 84.3% (107/127) and 94.7% (161/170) of 6-8 month and 9-24-month-olds seroconverted respectively.
Three decades later, investigators in Malawi found the same discrepancy between vaccination history and serosurvey results 27 . This 2015 study found poor sensitivity and positive predictive value (PPV, 75.8% and 79.9% respectively) for vaccination and infection history as predictors of positive measles serological tests.
Specificity was 10.3% and the negative predictive value (NPV) was also poor at 8.3 % 27 . Although the population in this more recent study was older and aged ≥18 months (median age 37 years), a 2001 study in the same setting reported the same discordance between seropositivity and vaccination history in 246 children <5 years. Here, 17% (4/23) and 67% (54/81) of children aged 8-12 and 12-23 months old were positive for measles antibody but the history of measles vaccination was 26% and 80% respectively 28 .
In a Gambian study, antibody test results following routine vaccination were compared against the vaccination records of 689 children aged 3 to 4 years. Vaccination records were retrieved from the Gambia Hepatitis Intervention Study (GHIS) main database and missing data obtained from vaccine card records during follow-up home visits. They found seroconversion in 91% (608/665) of the children with history of receiving the measles vaccine, 25% (14/57) of children without antibodies had been vaccinated before the age of 9 months and 7 of 24 children who did not receive the vaccine acquired measles antibodies possibly through natural infection 19 . The authors excluded children with incomplete vaccination histories either because they could not be traced for follow-up home visits or their immunization card were lost.
Ogunmekan et al. 17 measured measles antibody titres in 204 Nigerian children aged 6 to 12 months pre-and post-vaccination. On follow-up testing 6-8 weeks later, only 53.2% (91/171) of children without pre-vaccination antibodies seroconverted. The low seroconversion rates were attributed to low potency vaccines. No history of measles was given for the 33 children who were sero-positive before vaccination. However, in a more recent Nigerian report, Fowotade et al. 18 reported a significantly higher post vaccination seroconversion of 68.8% (196/286) in children 9-12 months old.
Contrary to the results from Nigeria, a better concordance was observed between crude and effective coverage (98.8% (878/889) and 94% (836/889) respectively) in Tanzanian children aged 18 months to 5 years 25 .
The utility of non-invasive biological specimens specifically oral-fluid (OF) for determining measles vaccine coverage was investigated by Nigatu W. et al. 23 among 1928 children aged 9 months to 5 years (pre-vaccination) and 745 individuals aged 9 months to 19 years (post-vaccination). There were differences in age-groups sampled pre-and post-vaccination because the post-vaccination survey carried out 6 months after the measles campaign was aimed at assessing population immunity and age susceptibility to measles. The study team documented receipt of measles vaccines delivered routinely and/or by supplementary immunisation activities. Pre-vaccination measles seroprevalence was 49% (1507/3075) in vaccinated and 21% (147/700) in unvaccinated children. The post-vaccination prevalence of measles antibodies was 80% in the 0-4 and 5-9 year olds, increasing to 96% in the [15][16][17][18][19] year olds as shown in Table 1. However, sensitivity of OF for antibody testing declined from 95% in individuals with measles IgG ≥1000 mIU/ml to 75% in individuals with measles IgG ≤1000 mIU/ml. In another study conducted in rural Kenya that evaluated the performance of OF as a measure of the effectiveness of a measles vaccination campaign, the pre-campaign and post-campaign measles seroprevalence was 60% (520/866) and 87% (753/866) respectively with a 70% reduction in the prevalence of the susceptible population at 85% crude coverage 24 . In addition, the proportion of weakly positive individuals by the assay used increased from 35% pre-campaign to 54% post-campaign.

Polio vaccine
In the single study of polio vaccine in West Africa, investigators assessed the effect of supplementary oral polio virus (OPV) vaccine doses in Gambian children aged 3 to 4 years 19 . Blood samples were tested for antibodies to polio virus types 1 and 3 using the standard polio neutralization assay. Seroprevalence for OPV 1 and OPV 3 was 81.5% (22/27) and 89% (24/27) respectively in children who had 3 recorded vaccine doses.

Yellow Fever vaccine
A household-based sero-epidemiologic survey of all age groups was conducted in Sudan after a mass vaccination campaign in response to yellow fever (YF) epidemic 26 . The serosurvey was designed to describe the epidemiology of the outbreak and to measure YF vaccine coverage. All the 84 (96.6%) and 3 (3.4%) individuals who reported having received and not having received the vaccines during the campaign respectively, had positive serological results for YF IgM antibodies. Of the 3 individuals who had positive serologic results despite no history of vaccination, 2 reported previous YF infection and 1 reported a Chikungunya virus-like illness.

Tetanus vaccine
Tapia et al. 20 collected serum and oral fluid (OF) from Malian infants, toddlers and adults males without a history of tetanus  One star (*) is awarded to self-reported or less objectively measured outcomes and two (**) are awarded when outcomes are assessed by independent blind observers or with record linkage (see Supplementary File, SF1).

Discussion
This review confirms the limited use of serosurveys in assessments of or to complement estimates of VC in sSA. We find evidence for the utility of serological data for determinations of effective vaccination coverage, despite their inability to distinguish population immunity induced by vaccines from those resulting from natural infection.
This review also highlights significant gaps between crude and effective VC for many vaccines currently in use as reported elsewhere 23,27,29,30 . Clearly, crude VC does not correctly identify all the non-or partially immunised disease susceptible population. For example, an effective coverage of >95% is required to induce sufficiently high herd immunity to interrupt transmission of measles and this is inferred from the crude VC typically derived from administrative records and/or parental recall 29,31 . There are several reasons why there are persistent outbreaks of VPDs such as measles in sSA. There is poor coverage resulting from poor access to and/or utilization of immunization services. In addition, there are evidence of cold chain failures leading to the delivery of non-viable/less potent vaccines to those with vaccination records 31,32 . Without significant improvements to the cold chain and crude coverage, many countries in the sub-continent will not achieve global and regional vaccination targets required to interrupt transmission of many VPDs.
The lower seroconversion rates for measles vaccine in children <9 months old may be attributable to interference by maternal antibodies 22 . Lower seroconversion reduces protection postvaccination. Since maternal antibodies wane over a period of 6-12 month 33,34 , a 2-dose schedule that delivers and achieves VC ≥95% for a measles-containing vaccine later in childhood (from the 2 nd year of life) is now deemed essential to on-going control efforts 35 . Because most cases of measles are subclinical, it was not surprising to find that a history of prior contact with or measles illness was poorly predictive of serological test results 36 .
Just like for measles vaccine, reported vaccination and previous infection were poor predictors of protective immunity for oral polio. The trivalent oral poliovirus vaccine (tOPV), a combination of all 3 types of polioviruses, is the most commonly used polio vaccine in routine childhood immunization 37 . Typically, tOPV is administered at 6 weeks or 1 st month of life at 4 week-intervals along with the DPT vaccine. Some countries administer an additional dose of tOPV at birth. TOPV and other preparations (monovalent or bivalent formulations containing types 1 and/or 3) are now used in campaigns to supplement routine childhood immunization 11 . The recent introduction of a single dose of injectable polio vaccine as part of the transition to replacing tOPV with bivalent OPV (bOPV, types 1 and 3) may lead to changing serological picture for polio in Africa. Therefore, serosurveys for polio may have value in countries in the pre-elimination phase of the disease 2 .
Unlike for measles, polio and yellow fever where the immune response induced by the vaccine is indistinguishable from that due to natural infection, 9 the presence of the antibody to tetanus toxin is usually, with rare exceptions, the result of immunization and not natural infection or subclinical exposure 38 . Therefore, serological tests for tetanus antitoxin could prove very helpful in ascertaining effective vaccination coverage for tetanus vaccine and as a surrogate for other co-administered vaccines. Although Tapia et al, showed a significant difference between tetanus antitoxin titres and receipt of the vaccine, these serological tests have not yet been sufficiently investigated to demonstrate the desired dose-response relationship between target markers and number of vaccine doses received 39 . Hepatitis B is thought to offer the same advantage as the testing of multiple antigens may help distinguish between a positive result due to natural infection and one following vaccination 40 . Despite the use of Hepatitis B seroprevalence data for determining the burden of disease and making recommendations for vaccination for several decades now 41 , none of the studies in this review investigated Hepatitis B.
Serological surveys may underestimate effective vaccination coverage due to waning immunity following vaccination as antibodies decay over time to concentrations that may be below the threshold for detection of currently available tests 9 . The risk for this is higher in older children and adults. However, most studies included in this review were conducted in the context of pre-and post-vaccination campaign evaluations so they involved mostly younger age groups (6-72 months) and occurred shortly after vaccination.
One important limitation of serological surveys is the risk of or low participation because of the requirement of invasive samples e.g. venous and capillary blood for testing. Unsurprisingly, this has stimulated an interest in the use of less invasive samples such as OF 42-44 and dried blood spots 45,46 . The use of oral fluid has been proposed as an alternative to overcome the risk of low participation due to venepuncture 9 . However, it appears the sensitivity of OF for measles antibody testing may vary according to serum measles IgG concentration 23 . In addition, serological assays of OF may be less sensitive in individuals with vaccine-induced immunity 23,47 . This is probably related to the serum antibody concentrations which are typically higher when induced by natural infection. The observed lower NPV for this specimen also suggests that the effectiveness of the vaccination programmes may be under-estimated by use of oral fluid assays 20 . Other evaluations of OF assays in Ethiopia and Kenya have yielded good results. However, the study in Kenya had no results from serum samples for comparison unlike the Ethiopian study that found a tight correlation between antibody prevalence in oral fluid and seroprevalence for measles for all age groups studied and for rubella in age groups <20 years 24,48 . Although these two studies highlight the potential for use of oral fluid as a replacement for serum in antibody prevalence surveys, this is an area that requires further investigation. This is important in the light of other studies outside the region that report poor sensitivity of OF and the deleterious impact of environmental conditions such as temperature and sample collection techniques on results from OF samples 29,42 . Although there is a longer history of using dried blood spots (DBS) as an alternative to venepuncture for serodiagnosis of many infectious diseases 49 , technical difficulties-eluting serum from DBS, the presence of haemoglobin and other debris that increase background reads in ELISA assays and cause lower sensitivity-have made them unpopular for vaccine serology 1 .
Our review had a few limitations. Many of the included studies did not justify their sample size, had a risk of residual selection bias in common (mainly because participants were not always randomly selected or the selection process was not well described) and ascertainment of vaccination was not always record-based (risk of ascertainment and recall bias). Finally, we may have missed important articles published in French and/or Portuguese which are also relevant to sSA.

Conclusion
The use of serological data has the potential to improve ascertainment of vaccination history in sSA. Serological tests of antitetanus antibodies may prove to be a reliable surrogate marker of vaccination because they are not confounded by naturally induced antibodies. There are knowledge gaps in the use of serological surveys for tetanus and hepatitis B and whether they are likely to show a dose-response relationship to the number of vaccine doses received. This is an important area of future research. Finally, more investigations to assess and optimize the performance of non-invasive samples such as OF and DBS for vaccine serosurveys are also required.

Data availability
All data underlying the results are available as part of the article and no additional source data are required.

Competing interests
No competing interests were disclosed. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Grant information
authors report sensitivity, specificity and positive and negative predictive values, they need to make clear what is the gold standard and what is being compared. Several studies appear to be vaccine immunogenicity studies (e.g. the measles study from The Gambia for which they report seroconversion, and the study from Nigeria). Vaccine immunogenicity studies are different than studies assessing population immunity through serological surveys.
The authors included serosurveys that used oral fluid rather than blood samples, but this review does not provide a sufficient evidence to assess the advantages and limitations of oral fluid samples.

5.
Dried blood spots are a method of storing blood samples not a method of collecting blood samples. One could perform venepuncture and spot the whole blood as dried blood spots. It is mistaken to refer to dried blood spots as a less invasive means of collecting blood or as an alternative to venepuncture, although frequently finger prick blood collection is used. I disagree that dried blood spots are "unpopular" for serological studies.

6.
The review does not address the assays used to measure seroprevalence nor the challenges and cost of conducting serological surveys in sub-Saharan Africa.

Minor comments
The statement that most cases of measles are subclinical is not correct. 1.
The word "hepatitis" in "hepatitis B" should not be capitalized.

2.
The authors need to make clear why they used smallpox scar as a surrogate marker of measles vaccination in the study by Breman. For this study, I suggest reporting the proportions seropositive rather than seronegative to be consistent with the rest of the text.

3.
For the measles study in Tanzania, I do not think it appropriate to refer to the seroprevalence results as "effective vaccine coverage" in a setting with potential wild-type measles virus transmission.

4.
For the yellow fever serosurvey in Sudan, the authors should more clearly describe how the serosurvey was used to both describe the epidemiology of the outbreak and measure yellow fever vaccine coverage at the same time.

5.
I would be cautious in using the term "effective coverage" in Table 1 for diseases other than tetanus unless it was certain wild-type virus was eliminated. "Seroprevalence" would be more accurate.

6.
The authors should be more specific when they state with regard to hepatitis B: "as the testing of multiple antigens may help distinguish between a positive result due to natural infection and one following vaccination". What antigens and what antibodies?

7.
Are the rationale for, and objectives of, the Systematic Review clearly stated? Partly For the studies included in the review, the authors are able to make a small number of conclusions as to why seroprevalence studies may correlate poorly with vaccine coverage. Many of these aspects are not unique to Africa and I wonder if the authors could comment more about which aspects may be generalizable to different settings and regions.
One minor comment is that haemophilus influenzae is spelt without the "e".
Are the rationale for, and objectives of, the Systematic Review clearly stated? Partly

Is the statistical analysis and its interpretation appropriate? Yes
Are the conclusions drawn adequately supported by the results presented in the review? Yes publications that should have been included given the defined objectives of the authors.

Major comments:
1. At the time the paper was published online without review (23Feb2018), the search was already 13 months out of date (last search 30Jan2017). Is there any reason why it was not feasible to update the search again closer to the publication date? Given the rapidly evolving topic area, I suggest including 2017 and 2018 articles up to publication time, if possible. . I'm surprised that in Figure 1 only three articles were excluded for being serosurveys not including vaccination coverage. I would guess that the majority of published serosurveys in Africa did not collect vaccination coverage (e.g., rubella serosurveys of adult women). Other articles may not report coverage in the abstract but do report it in the full text. An example would be Aboud et al., 2000 13 . Inability to capture these articles should be a stated limitation of your review. If you re-categorized the 1031 articles that were "non-relevant" into smaller groups, and the number excluded because of not including vaccination coverage was high, you may be able to recommend inclusion of vaccination status in serosurveys where possible to inform interpretation of results and provide better information for the program.
Minor comments: 1. MacNeil et al 2014 14 summarizes the limitations of using serosurveys to assess vaccination coverage. In my experience, the limitations are not "well known" as the authors state in the introduction. A major result described in the abstract is discordance of coverage and serosurvey results for diseases where natural infection causes immunity. This limitation of serosurveys (inability to discriminate immunity from natural infection and vaccination, except for tetanus) seems worth describing in the introduction.
2. Another limitation of serosurveys that is not mentioned is misclassification bias related to use of non-gold standard tests, e.g., ELISAs with poor sensitivity and poor specificity. This issue is also discussed in MacNeil et al 2014 14  Are the rationale for, and objectives of, the Systematic Review clearly stated? Yes Are sufficient details of the methods and analysis provided to allow replication by others? Partly

Is the statistical analysis and its interpretation appropriate? Yes
Are the conclusions drawn adequately supported by the results presented in the review? Yes