HPV infection and pre-term birth: a data-linkage study using Scottish Health Data

Background: We aimed to investigate whether infection with high-risk (HR) types of human papilloma virus (HPV) or HPV-associated cervical disease were associated with preterm birth (<37 weeks gestation). In a sub-group of younger women who were eligible for the HPV vaccine, we aimed to determine whether prior vaccination against the specific HPV-types, HPV-16 and -18 modified preterm birth risk. Methods: This was a data-linkage study, which linked HPV-associated viral and pathological information (from the Scottish HPV Archive) from women aged 16-45 years to routinely collected NHS maternity- and hospital-admission records from 1999–2015. Pregnancy outcomes from 5,598 women with term live birth (≥37 weeks gestation, n=4,942), preterm birth (<37 weeks gestation, n=386) or early miscarriage (<13 weeks gestation, n=270). Of these, data from HPV vaccine-eligible women (n=3,611, aged 16-25 years) were available, of whom 588 had been vaccinated. HPV-associated disease status was defined as: HR HPV-positive no disease, low-grade abnormalities or high-grade disease. Results: High-grade HPV-associated cervical disease was associated with preterm birth (odds ratio=1.843 [95% confidence interval 1.101–3.083], p=0.020) in adjusted binary logistic regression analysis, in all women, but there were no associations with HR HPV-infection alone or with low-grade abnormalities. No associations between any HPV parameter and preterm birth were seen in vaccine-eligible women, nor was there any effect of prior vaccination. Conclusions: HPV-associated high-grade cervical disease was associated with preterm birth, but there were no associations with HR HPV-infection or low-grade cervical disease. Thus HPV-infection alone (in the absence of cervical disease) does not appear to be an independent risk factor for preterm birth. For women who have undergone treatment for CIN and become pregnant, these results demonstrate the need to monitor for signs of preterm birth.


Introduction
Human papillomavirus (HPV) is a sexually transmissible virus that infects cervical cells. Normally, the virus is cleared by the immune system, becoming undetectable within two years 1 . Persistent HPV infection causes pre-cancerous cervical intraepithelial neoplasia (CIN); left untreated, CIN can develop into cervical cancer 2 . Over 200 HPV types exist, of which 14 are oncogenic or high-risk (HR). Two such HR types, HPV-16 and -18, are responsible for 70-80% of invasive cervical cancers 3 . Treatment of CIN and cervical cancer is by removal of the abnormal cervical cells 4 .
The cervix is important for maintenance of pregnancy, with barrier and immune mechanisms that protect the growing foetus 5 . HPV-infection may alter cervical function, possibly increasing the risk of intrauterine infections 6 and subsequent complications such as preterm birth 7 or miscarriage 8 .
The relationships between HPV infection and pregnancy outcomes are unclear. Some studies have shown no association 9,10 , whereas others have suggested that HPV-infection, cervical disease and/or its treatment are associated with adverse pregnancy outcomes, such as miscarriage 11 , premature preterm rupture of membranes (PPROM) 12 or spontaneous preterm birth 13, 14 .
We performed a data-linkage cohort study, linking women with HPV viral and pathological data with their own pregnancy records. The aims for this study were to determine whether HR HPV-infection, specific HR HPV types, or the presence of HPV-associated cervical disease were associated with preterm birth (<37 weeks gestation), early miscarriage (<13 weeks gestation) or stillbirth (pregnancy loss at ≥24 weeks gestation) and whether previous vaccination against HPV-16/18 affected these outcomes.

Study background
This study was a data-linkage study, linking data from women with HPV viral and pathological data contained within the Scottish HPV Archive (see below) with their own pregnancy records. After data-cleaning, the retrospective data from the women in the resulting cohort were analysed for associations between the HPV viral/pathology parameters and adverse pregnancy outcomes: preterm birth (<37 weeks gestation), early miscarriage (<13 weeks gestation) and stillbirth (≥24 weeks gestation). The study population for whom data were obtained and linked were women who had samples and data within the Scottish HPV Archive and who had had details of pregnancy outcomes.

Eligibility criteria
Women whose HPV-infection status was known at the time of birth were included. As HPV is usually cleared within 2 years 1 and women are advised to delay their next scheduled cervical smear until at least 12 weeks postnatally 4 , women who had a pregnancy up to 1 year prior to the cytology date were also included. In women who had had more than one pregnancy, the pregnancy outcome nearest the cytology date was included.
Exclusion criteria for the study were: Multiple (twin) pregnancies, as these babies are more often routinely delivered preterm; therapeutic terminations, ectopic pregnancies and trophoblastic disease were also excluded.

Definition of terms
The definitions for pregnancy outcomes were: Term live births were women who had a live birth at term (≥37 weeks gestation); All preterm births were women with births before 37 weeks' gestation including miscarriages of ≥13 weeks gestation; Spontaneous preterm births were women with births prior to 37 weeks' of gestation, including those with PPROM and miscarriages of ≥13 weeks gestation, but excluding iatrogenic preterm births (those delivered by elective caesarean section or those which were induced); Early Miscarriage was any spontaneous pregnancy loss before 13 weeks' gestation; Stillbirth was defined as any pregnancy loss after 24 weeks' gestation.
Sources of patient information used for the data-linkage Scottish HPV Archive. Scotland has a cervical screening programme with population coverage of around 70% 15 . Until June 2016, women aged 20-60 years were screened; thereafter the age-range was 25-65 years. In September 2008, a routine, school-based HPV vaccination programme for girls aged 11-13 years was introduced, with a "catch-up" immunisation programme for girls aged 13-17 years 16 . In 2009, the Scottish HPV Archive was established as a biobank for HPV-associated research. It is a collection of collections which includes residual cervical samples from women attending routine cervical screening and colposcopy clinics, from the year 2000 to the present. Many samples within the Archive have a patient identifier (community health index [CHI] number) and associated clinical, cytology and histology results, obtained from the Scottish Cervical Cytology Recall System (SCCRS). SCCRS is the national IT system that supports the Scottish cervical screening program and contains a woman's entire screening record 4 . A number of samples in the Archive are annotated with HPV status as a consequence of immunisation surveillance and specific research projects 17,18 , using different HPV assays depending on the nature and objectives of the research projects (  19 . Audit of SMR01 found that accuracy of clinical data was more than 89%, when compared with written medical records 22 . Miscarriage data were obtained from both SMR01 and SMR02. If an SMR01 record had no corresponding SMR02 record, we assumed that the miscarriage occurred before 13 weeks' gestation, as it is usual practice for pregnancies to be registered on SMR02 before 13 weeks gestation 23 . Scottish birth statistics from the Scottish population, obtained from SMR02 and SMR01, are available from the NHS NSS ISD website, Births in Scottish Hospitals 23 .
Miscarriage may occur before the mother is aware she is pregnant, and is often managed in an outpatient or general practice setting 23

Data extraction
The data-linkage and -cleaning processes are summarised in Figure 1. Data were transferred between the Scottish HPV Archive and eDRIS, where data-linkage was performed. From all the women's records within the Scottish HPV Archive (N=31,320), those who had had pregnancies recorded in SMR02, SMR01 and NRS were identified by the CHI number. Coded identification numbers of women with pregnancy records were sent back to the Scottish HPV Archive (n=10,572), from where HPV-genotype, cytology date and result, histology date and result, vaccination date and dosage were collated. These data are denoted as 'HPV viral/pathology data'. The HPV viral/pathology data and corresponding SMR02, SMR01 and NRS records were collated, anonymised and placed in the Safe Haven. For data cleaning and analysis, only the Safe Haven and the linked datasets were accessed.
Pregnancy admission records from SMR02, SMR01 and NRS records for women in the Scottish HPV Archive were received from January 1981 to August 2015. Many of the older women within the Scottish HPV Archive had their pregnancies before their HPV-infection status was known and so were excluded. Records in the Scottish HPV Archive started in 2000, so pregnancy admissions before 1999 were removed; therefore, the time-period for this study was 1999-2015. Duplicate records were removed. Information from multiple admissions per pregnancy outcome were condensed into one record. The variables obtained from SMR01, SMR02, NRS records and the Scottish HPV Archive are presented in Table 1, together with details of data coding and cleaning. Pregnancy records were coded without reference to HPV-infection status. The proportions of women with specific pregnancy outcomes within our final cohort was compared with that of the Scottish population 23 to determine how representative our cohort was of the Scottish population.
Classification of individuals within the dataset HPV-infection status of each woman was defined as HR HPVpositive if there was a positive result for at least one HR HPV type (Table 1). Infections with a mixture of HPV types were common. Low-risk HPV-positive samples (Table 1) and HPVnegative results were considered HR HPV negative.
Women were classed as HPV-16/18-positive if they had samples containing HPV-16 and/or HPV-18, irrespective of other types present. Women were classed as non-16/non-18 HR HPV-positive if they had samples that were negative for HPV-16 or HPV-18 but contained other known HR HPV types.

HPV vaccination status.
Some women had received the bivalent HPV vaccine 17 (1-3 doses) against HPV-16 and -18. Vaccineeligible women were classified as: those who had been vaccinated, those eligible (by age) but unvaccinated, or those eligible but whose vaccination status was unavailable; women born before 1990 were too old to have been eligible for vaccination.
Scottish National guidelines require over 90% of women with HPV-associated high-grade cervical disease to have undergone excisional treatment 4 , but data for the specific treatments used were not available.

Determining cohort reliability
To determine how representative our cohort was of the Scottish population, the proportions of women with specific HPV viral/pathology parameters within our final cohort, were compared with a recent HPV prevalence study in women attending screening in Scotland 28 , as HPV results are not recorded as part of routine cervical screening. Pathology data (low-and high-grade disease) were compared with data from ISD 29 .
Potential confounders for pregnancy outcomes were obtained from SMR02 and SMR01: ethnicity, parity, Scottish Index of Multiple Deprivation (SIMD), smoking history and during current pregnancy, diabetes (pre-existing and gestational) and hypertension (any). Coding and definitions for these variables are shown in Table 1. Continuous variables obtained were maternal age and maternal BMI. Neither variable was normally distributed, nor was any transformation able to normalise the data. Maternal age (range 16-45 years) was severely skewed, with 62.9% of women in our cohort aged <25 years. As we wished to investigate the effects of vaccination, and the vaccine-eligible women were aged <25 years, we chose 20 years as the age cut-off point for the analyses, comparing those aged <20 years with those >20 years. Maternal BMI data were categorised: underweight (BMI <19.9), normal (BMI 20.0-24.9), overweight (BMI 25.0-29.9) and obese (BMI ≥30.0).

Statistical analysis
All statistical analyses were carried out using SPSS v22 (© IBM Corporation, USA). Missing data analysis (Table 2) of the cleaned full dataset was carried out using the 'Multiple Imputation' function, which runs regression models of existing data to replace missing data: i.e. the programme looks at patterns of the data available and makes probability judgements (imputations) to give estimates of the missing data. Data was missing for Scottish Index of Material Deprivation (SIMD), ethnicity, parity, hypertensive disorder, current smoking, smoking history, diabetes and maternal BMI (Table 3). Variables were imputed together so that any potential interactions could be taken into account 30 . For the logistic regression analyses, pooled results from five imputations are presented in the results. The HPV viral/pathology parameters (HR HPV, HPV-16/18 genotype, HPV-associated cervical disease and HPV vaccination status) were not imputed, as these were the primary predictive factors of interest for the analyses.
The proportions of maternal characteristics according to pregnancy outcomes or according to HPV viral/pathology parameters were compared by chi squared or Fisher's exact test analysis and are shown in Table 4 and Table 5, respectively. Binary logistic regression analysis was used for each pregnancy outcome, with women who had had term live births as the comparator group and each HPV viral/pathology parameter as the predictive factor under analysis. Odds ratios (OR) and 95% confidence intervals (CIs) are presented from unadjusted and adjusted models in all women. To investigate the effects of prior HPV vaccination, the models were run in vaccine-eligible women, with and without vaccination status included. Adjusted models used the imputed dataset and adjusted for maternal factors that might affect pregnancy outcomes: ethnicity, SIMD, parity, age, BMI, diabetes, hypertension and current smoking.

Cohort screening, selection and characteristics
Pregnancy admission records (N=32,520) were received for women who had been pregnant and had HPV viral/pathology data in the Scottish HPV Archive (N=10,572). After datacleaning and exclusions, we had obtained 7,512 pregnancy Imputed data are the pooled results of five imputations. The numbers and percentages presented are representative of the population as a whole and represent the estimated values that would have been obtained had the dataset been complete. Imputed data were used in the logistic regression models. ‡ Maternal age was complete. Age range of women in each group is shown. || Scottish Index of Material Deprivation (SIMD) quintiles were combined as most deprived (quintiles 1 &2) vs. least deprived (quintiles 3-5). ¶ Body mass index (BMI) was calculated as weight (kg) divided by the square of the patient's height (m 2 ) # Diabetes includes gestational diabetes and pre-existing diabetes (Table 1). **Hypertension did not distinguish between pre-existing or gestational hypertension within SMR02 (Table 1)  outcomes from 5,626 women ( Figure 1). Of these, 32 were stillbirths, which, although these had been pre-defined as one of the pregnancy outcomes for investigation, were too few for any meaningful analysis and were excluded. For the women who had had more than one pregnancy, we chose the pregnancy nearest the cytology date. The final cohort consisted of 5,598 women.

Table 5. Maternal characteristics and HPV viral/pathology parameters in the cohort of women.
Women had had term live births (≥37 weeks gestation, n=4942, 88.3%); all preterm birth (<37 weeks gestation, n=386, 6.9% [including those who had a late miscarriage, n=26]) and early miscarriage (<13 weeks gestation, n=270, 4.8% [obtained from SMR02, n=156, and SMR01, n=114]). On comparison with Scottish birth statistics 23 , the age distribution of the cohort was much younger than that of the Scottish childbearing population. There were more pregnancy outcomes in women of <25 years (62.9%) than in the Scottish population 23 (26.5%) and a smaller proportion of mothers aged 25-34 years (28.8% vs. 53%) or ≥35 years (8.3% vs. 20.5%) in our cohort, respectively. The percentage of all women who had a preterm birth in our cohort (6.9%) was similar to that of the Scottish population 23 , which ranged between 5.8 and 6.7% (between 1999 and 2015). The percentage of women having hospital admission for early miscarriage in our population was much lower (4.8%) than that in the Scottish birth data 23 . Due to this low rate of miscarriage, the miscarriage data were deemed to be problematic due to uncertainty in case ascertainment and was thought to not be representative of the Scottish child-bearing population. Therefore, the results for early miscarriage are not presented.
The characteristics of the women according to the HPV viral/ pathology parameters are shown in Table 5 for the whole cohort. The percentage of women with HR HPV-positive results in our cohort was 27.5%, which is higher than the HR-HPV prevalence (18%) seen previously in the Scottish screening population 28 , probably due to the nature of 'selective' sampling in the Scottish HPV Archive. Similarly, the percentage of women with HPV-16/18 was higher than that in the routine prevalence in the same population (9.7% vs. 2.9%, respectively). For HPVassociated disease, some women were HR HPV-negative or of unknown genotype but showed evidence of HPV-associated disease. The proportion of women with low-grade disease was higher in our cohort than in the national screening data (16.0% vs. 7.5%, respectively) and our cohort had a higher percentage of women with high-grade cervical disease (2.9% vs. 1%, respectively).

Analysis of the cohort
The numbers and percentages for the maternal factors were compared by pregnancy outcomes (term live births vs. all preterm birth, Table 4). There were no significant differences in maternal age, ethnicity, parity, SIMD (most vs. least deprived), BMI or hypertension between women who had term live births compared with those who had preterm births. A higher proportion of all women who had a preterm birth were current smokers when compared with those who had term live births (P<0.001). Similarly, a higher proportion of all women who had a preterm birth also had diabetes than those who had term live birth (P=0.063, Table 4).
The numbers and percentages of women in each analysis and the unadjusted and adjusted OR from binary logistic regression are shown for each HPV viral/pathology parameter and all women who had a preterm birth, for all women (Table 6) and vaccine-eligible women (Table 7); similarly results for spontaneous preterm birth are shown in Table 8 and Table 9.
HR HPV. Binary logistic regression showed that women who carried any HR HPV had increased odds of having a preterm birth in unadjusted models, but the association was lost in adjusted models (Table 6). In vaccine-eligible women there was no association between HR HPV and all preterm birth and no effect of vaccine (Table 7). Similarly, for spontaneous preterm birth, there was no association between HR HPV and spontaneous preterm birth in the whole cohort (Table 8), nor in vaccine-eligible women and no effect of vaccine (Table 9). Current smoking and/or diabetes were independently associated with increased odds of all preterm birth and/or spontaneous preterm birth. Hypertension was an independent predictor of all preterm birth in all women but not vaccine-eligible women.

HR HPV 16/18 types.
Only 970 women had known specific HR HPV types. There were no associations between carriage of HPV16/18 types or non16/non18 HPV types and all preterm birth (Table 6 and Table 7) or spontaneous preterm birth (Table 8 and Table 9) in the whole cohort nor in vaccineeligible women, and vaccination had no effect. Current smoking and/or diabetes were independently associated with increased odds of any and/or spontaneous preterm birth. Hypertension was an independent predictor of any preterm birth in all women but not vaccine-eligible women.

HPV-associated cervical disease.
In the whole cohort of women, high-grade cervical disease was significantly associated with ~80% increased odds of any and spontaneous preterm birth (Table 6 and Table 8, respectively) in unadjusted and adjusted models of all women. In the vaccine-eligible women there were no associations between any level of HPV-associated disease and no effect of vaccination in any preterm birth or spontaneous preterm birth (Table 7 and Table 9, respectively). Current smoking and diabetes were also associated with an increased odds of any or spontaneous preterm birth in all women. Younger maternal age showed a decreased association with HPV-associated cervical disease in the models for both any and spontaneous preterm birth.

Discussion
This study found associations between different HPV viral/ pathology parameters and adverse pregnancy outcomes. Specifically, high-grade HPV-associated cervical disease was associated with preterm birth and spontaneous preterm birth in the whole cohort women, but not in the vaccine-eligible subgroup of younger women. No associations were seen between HR HPVpositive (with no disease) or low-grade HPV-associated cervical disease and all or spontaneous preterm birth.
Associations between high-grade cervical disease and all preterm birth or spontaneous preterm birth have been suggested in Table 6. Any preterm birth compared with term live-birth in all women, according to HPV viral/pathological parameters. Table shows the pregnancy outcomes: term live births (>37 weeks gestation) and any preterm birth (<37 weeks gestation) compared by each HPV parameter for all women in the cohort. The numbers and percentages in each group are shown. The odds ratios (OR) and P values from unadjusted and adjusted binary logistic regression models are shown. Significant associations are highlighted in bold.  1.135-3.125) 0.014  1.843 (1.101-3.083) 0.020 * Due to missing HPV data, numbers and percentages relate to the numbers that were present in each analysis and do not add up to the total numbers of women in each pregnancy outcome. † Adjusted binary logistic regression models used the imputed dataset and are pooled results from 5 imputations. Models were adjusted for ethnicity, SIMD (most deprived vs. least deprived), maternal age, parity, smoker in current pregnancy, diabetes, hypertensive disorder and maternal BMI. ‡ Criteria for HPV-associated cervical disease are shown in the methods and Table 1.

HPV Parameter
Cofactors that had a significant association with preterm birth from these models were: previous studies 13, 14 . In this study, such associations were not seen in the younger, vaccine-eligible subgroup of women, suggesting that the older women were infected for longer and had time to develop high-grade cervical disease 30 . Previous studies showing associations between HPV-associated cervical disease and preterm birth have been ambiguous regarding whether these were due to HPV-infection itself, development of associated CIN lesions or the excision of these lesions 31,32 . Low-grade cervical lesions are more likely to be associated with productive HPV-infections, where the viral life-cycle is completed, compared with infections associated with severe disease where the viral life-cycle is poorly supported 33 . Productive infections are also associated with a peak/burst of viral load over a narrower time-frame than infections associated with high-grade disease 34 . Therefore, in the absence of any associations with HR HPVinfection alone or low-grade cervical disease, and given that over 90% of those who had high-grade disease would have been treated 4 , we suggest that the association with high-grade cervical disease, in our cohort, is likely to be due to the treatment of HPV-associated cervical disease rather than a direct effect of HPV-infection itself. The clinical implication is that women who have undergone treatment for HPV-associated cervical disease should be closely monitored in subsequent pregnancies for any signs of preterm birth.
Although only HPV-associated high-grade cervical disease was associated with all preterm birth and spontaneous preterm birth, current smoking was consistently associated with spontaneous preterm birth. Smoking is a previously recognised risk factor for spontaneous preterm birth 35 . Diabetes and hypertension were also independently associated with all preterm birth, but this was not surprising as these are likely to be the indicative reasons for early delivery of the baby. A recent study suggested that HPV infection in the placenta was associated with a number of pregnancy complications 36 , including gestational diabetes. However, our data did not find an increased associa- * Due to missing HPV data, numbers and percentages relate to the numbers that were present in each analysis and do not add up to the total numbers in each pregnancy outcome. † Adjusted binary logistic regression models used the imputed dataset and are pooled results from 5 imputations. Models were adjusted for ethnicity, SIMD (most deprived vs. least deprived), maternal age, parity, smoker in current pregnancy, diabetes, hypertensive disorder and maternal BMI. Any cofactors that had a significant association with preterm birth from these models are shown. ‡ Criteria for HPV-associated cervical disease are shown in the methods.
Cofactors that had a significant association with preterm birth from these models were:  * Due to missing HPV data, numbers and percentages relate to the numbers that were present in each analysis and do not add up to the total numbers in each pregnancy outcome. † Adjusted binary logistic regression models used the imputed dataset and are pooled results from 5 imputations. Models were adjusted for ethnicity, SIMD (most deprived vs. least deprived), maternal age, parity, smoker in current pregnancy, diabetes, hypertensive disorder and maternal BMI. Any cofactors that had a significant association with preterm birth from these models are shown. ‡ Criteria for HPV-associated cervical disease are shown in the methods. § Where there were groups containing <10 women, the numbers were not released from the Safe Haven and only the percentages are given.
Cofactors that had a significant association with preterm birth from these models were: tion between HPV viral/pathology data and diabetes and so do not support such an association.
There are strengths and weaknesses for this study. The strengths of this study are that this was a data linkage study starting with women for whom HPV-infection status was known. The use of the Scottish HPV Archive meant that the HPV viral/pathology parameters for these women had been robustly characterised and the HPV information was not available from any other source. The use of routinely collected data should also reduce bias in the recording of clinical information. Primary HPV screening in Scotland will soon include results from HPV testing. Such information may facilitate a future population-wide analysis of the effects of HPV-infection and pregnancy outcomes in the Scottish population.
A weakness of the study is that we did not have details of any treatments for cervical disease, so we have had to rely on an assumption that women with high-grade disease received excisional treatment as per national guidelines. The age-distribution of the women did not reflect that of the Scottish childbearing population 23 , and therefore, it is hard to determine how representative these results are of the wider population. The study cohort was drawn from the general population by the fact that they underwent cervical screening, but were considerably younger at pregnancy outcome and had higher proportions of women with HPV-infection and HPV-associated cervical disease, due to the selective nature of the samples within the Scottish HPV Archive. The preterm birth rates were similar to that in the Scottish population. The early miscarriage data were obtained from maternity/hospital admission records, but would not include miscarriages managed at home, through outpatient and emergency departments or by General Practitioners. The risk of miscarriage increases with age, particularly after the age of 40 years, although a slightly increased risk of miscarriage exists in women aged under 20 years 38 . The miscarriage rate in our population was much lower than the that reported for miscarriages managed in hospitals 23 . While this lower rate may particularly reflect the youth of our cohort, the differences are hard to explain, and so these data were not presented.
In summary, our data linkage study has found no evidence that HPV infection per se, or low-grade HPV-associated cervical disease was associated with preterm birth. High-grade disease, the majority of which is treated with excisional treatments, was associated with spontaneous preterm birth. This is consistent with previous evidence linking excisional cervical treatment with increased risk of spontaneous preterm birth 13 .

Data availability
Underlying data The linked dataset was analysed within the NHS National Safe Haven, provided by NHS Research Scotland. The Safe Haven is a remote server through which the researcher accesses the health data and services to enable research while protecting the confidentiality of the data. Data remains under the control of the NHS and complies with legislative and NHS policies. The linked dataset is archived within the Safe Haven and is available by application to NHS Scotland via the electronic Data Research and Innovation Service (eDRIS), within NHS National Services Scotland (https://www.isdscotland.org/Products-and-Services/ eDRIS/).
To apply for access to these data, please read the guide for researchers and then complete the Enquiry Form, describing your planned study and the data required, and email it to nss.edris@nhs. net.

Grant information
This work was funded by Tommy's, the Baby Charity. The work was carried out at the MRC Centre for Reproductive Health which is funded by MRC Centre Grant (MR/N022556/1). SJS is funded by a Wellcome Trust Clinical Career Development Fellowship (209560).
The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.