You are here

  • Home
  • Archive
  • Volume 13, Issue 4
  • Assessing the effects of duration of birth interval on adverse pregnancy outcomes in sub-Saharan Africa: a propensity score-matched analysis

Article Text

Article menu
Epidemiology
Original research
Assessing the effects of duration of birth interval on adverse pregnancy outcomes in sub-Saharan Africa: a propensity score-matched analysis
  1. Getayeneh Antehunegn Tesema1,
  2. Achamyeleh Birhanu Teshale1,
  3. Yigizie Yeshaw2,
  4. Dessie Abebaw Angaw3,
  5. Ayenew Lakew Molla1
  1. 1Department of Epidemiology and Biostatistics, University of Gondar College of Medicine and Health Sciences, Gondar, Ethiopia
  2. 2Medical Physiology, University of Gondar College of Medicine and Health Sciences, Gondar, Ethiopia
  3. 3Department of Epidemiology and Biostatistics, College of Medicine and Health Sciences, Institute of Public Health, Gondar University, Gondar, Ethiopia
  1. Correspondence to Getayeneh Antehunegn Tesema; getayenehantehunegn{at}gmail.com

Abstract

Objectives Unlike high-income countries, sub-Saharan African countries have the highest burden of adverse pregnancy outcomes such as abortion, stillbirth, low birth weight and preterm births. The WHO set optimal birth spacing as a key strategy to improve pregnancy outcomes. Estimating the impact of short and long birth intervals on adverse pregnancy outcomes based on an observational study like the Demographic and Health Survey (DHS) is prone to selection bias. Therefore, we used the propensity score-matched (PSM) analysis to estimate the actual impact of short and long birth intervals on adverse pregnancy outcomes.

Design A community-based cross-sectional study was conducted based on the DHS data.

Setting We used the recent DHS data of 36 sub-Saharan African countries.

Participants A total of 302 580 pregnant women for stillbirth and abortion, 153 431 for birth weight and 115 556 births for preterm births were considered.

Primary outcome measures To estimate the impact of duration of birth interval (short/long) on adverse pregnancy outcomes, we used PSM analysis with logit model using psmatch2 ate STATA command to find average treatment effect on the population (ATE), treated and untreated. The quality of matching was assessed statistically and graphically. Sensitivity analysis was conducted to test the robustness of the PSM estimates using the Mantel-Haenszel test statistic.

Results The prevalence of short and long birth intervals in sub-Saharan Africa was 46.85% and 13.61%, respectively. The prevalence rates of abortion, stillbirth, low birth weight, macrosomia, and preterm births were 6.11%, 0.84%, 9.63%, 9.04%, and 4.87%, respectively. In the PSM analysis, the differences in ATE of short birth intervals on abortion, stillbirth, low birth weight, and preterm births were 0.5%, 0.1%, 0.2%, and 0.4%, respectively, and −2.6% for macrosomia. The difference in ATE among the treated group was 1%, 1%, and 1.1% increased risk of abortion, low birth weight, and preterm births, respectively, while there was no difference in risk of stillbirth between the treated and control groups. The ATEs of long birth intervals on abortion, stillbirth, low birth weight, macrosomia and preterm births were 1.3%, 0.4%, 1.0%, 3.4%, and 0.2%, respectively. The ATE on the treated group had 0.9%, 0.4%, 2.4%, 2.8%, and 0.2% increased risk of abortion, stillbirth, low birth weight, macrosomia, and preterm births, respectively. The estimates were insensitive to hidden bias and had a good quality of matching.

Conclusion Short and long birth intervals had a significant positive impact on stillbirth, abortion, low birth weight, macrosomia and preterm births after matching treated and control groups by observed variables. These findings highlighted maternal and newborn healthcare programmes and policies to empower reproductive-aged women to exercise optimal birth spacing to reduce the incidence of stillbirth, abortion, low birth weight, macrosomia and preterm births.

  • PUBLIC HEALTH
  • EPIDEMIOLOGY
  • OBSTETRICS
  • Reproductive medicine

Data availability statement

Data are available in a public, open access repository. Data are available in a public, open access repository. These data can be found from https://dhsprogram.com/data/dataset_admin/login_main.cfm?CFID=52406619&CFTOKEN=acf80f39c3b63be9-65910752-FB64-58E6-E6F9496A4F66E431.

http://creativecommons.org/licenses/by-nc/4.0/

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.

https://doi.org/10.1136/bmjopen-2022-062149

Statistics from Altmetric.com

    Request Permissions

    If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

    STRENGTHS AND LIMITATIONS OF THIS STUDY

    • The matching was done based on the observed variables only, and there is a possibility of residual confounding (unobserved variables).

    • Moreover, the Demographic and Health Survey (DHS) is a cross-sectional study and it is prone to social desirability and recall bias.

    • Despite the limitations, the study has strengths. First, this study is based on nationally representative DHS data with a high response rate.

    • Second, the DHS uses a standardised questionnaire for the data collection, which is consistent across all 36 countries.

    • Furthermore, this study is the adjustment for potential confounders using the propensity score-matched approach in the estimation of the association between short/long birth intervals and adverse pregnancy outcomes.

    Background

    Adverse pregnancy outcome is defined as a low birth weight (LBW) (birth weight <2500 g), macrosomia (birth weight >4000 g), stillbirth (fetal death at or after 7 months of gestation), abortion (termination of pregnancy before 7 months of gestation) and preterm birth (less than 37 weeks at birth).1 2 Globally, adverse pregnancy outcomes continue to be a serious public health problem specifically in low/middle-income countries like sub-Saharan African (SSA) countries.3–6 Worldwide, an estimated 13 million, 46 million, and 2.6 million pregnancies end in preterm birth, abortion, and stillbirth annually, respectively.7–9 Similarly, it is estimated that 15%–20% of all births are LBWs and 3%–15% of all pregnancies are macrosomia globally,10–12 which is associated with significant short-term and long-term consequences.13

    Many studies evidenced that birth interval length, the duration between two consecutive live births, as a significant predictor of adverse pregnancy outcomes14–16; after live birth, the recommended interval before attempting the next pregnancy is at least 24 months to reduce the risk of adverse maternal, perinatal and infant outcomes.17 18 This recommendation is based on the findings that intervals shorter than 33 months and longer than 60 months are associated with an elevated risk of adverse pregnancy outcomes.19–22 Short birth interval (SBI) and long birth interval (LBI) are more common among women in low/middle-income countries.23–25

    Birth spacing has a significant effect on pregnancy outcomes.26 Birth interval is the length of the interval between two consecutive births: numerous studies have shown that short birth spacing is significantly linked with an increased risk of abortion, stillbirth, preterm birth, macrosomia and LBW.27 28 These associations stem from the biological factor commonly referred to as maternal depletion syndrome.29 Closely spaced pregnancies may not allow sufficient time for the mother to restore her depleted micronutrient and macronutrient stores from the previous birthing, which in turn may reduce her ability to provide a favourable fetal growth environment in subsequent pregnancies and sufficient breast milk production post-delivery.30

    There is limited empirical evidence on the actual impact of preceding/subsequent birth intervals on adverse pregnancy outcomes. The impact of SBI/LBI on adverse pregnancy outcomes such as LBW, abortion, stillbirth, preterm births and macrosomia has been attributed to the biological effects related to the ‘maternal depletion syndrome’ or more generally the woman not fully recuperating from one pregnancy before supporting the next one.31 32

    As advocated by the WHO, optimal birth spacing is one of the most effective strategies to reduce adverse pregnancy outcomes in low/middle-income countries.23 33 Numerous studies evidenced the association between birth interval and adverse pregnancy outcomes.15 34 However, the SBI/LBI in and of itself may not directly result in adverse pregnancy outcomes, rather it might be the mothers may differ across known and unknown factors to influence adverse pregnancy outcomes. Therefore, traditionally, to control for such confounding, the association between SBI/LBI and adverse pregnancy outcomes in statistical analysis has been done via regression analysis. However, bias (residual confounding or hidden bias) persists, as the regression analysis adjusts for the observed variables only. For instance, the positive effect of SBI/LBI on adverse pregnancy outcomes might be biased because of the distribution of factors influencing birth intervals among mothers who had SBI/LBI to mothers who had optimal birth spacing, even when such factors are controlled for within the regression model. Besides, women who had SBI/LBI have more risks of adverse pregnancy outcomes compared with women who had optimal birth intervals (OBIs) because of unobserved characteristics, which might introduce bias.

    Therefore, propensity score matching (PSM) is an appropriate approach to estimating the actual impact of birth intervals on adverse pregnancy outcomes. PSM is a methodological technique that aimed to remove bias by matching treated (SBI/LBI) and untreated (OBI) pregnant mothers with similar conditional probability to receive the treatment (SBI). In this study, we matched the mothers with OBI/LBI to mothers with SBI with similar propensity score values for birth intervals. Then, it can be reasoned that any difference in adverse pregnancy outcomes is attributed to birth interval only. However, as to our search of the literature, there was no study done in SSA that used propensity scores to assess the impact of SBIs on adverse pregnancy outcomes. The present study addresses the methodical limitations of previous studies by examining the effect of SBI/LBI on adverse pregnancy outcomes in SSA using PSM analysis.

    Methods

    Study design and period

    A community-based cross-sectional study was conducted based on the Demographic and Health Survey (DHS) data of 36 SSA countries. The DHS is conducted every 5 years to generate updated health and health-related indicators in low/middle-income countries. Data on stillbirth, abortion and birth weight were collected in the 36 SSA countries but data on stillbirth were available in 11 SSA countries only (Angola, Benin, Burundi, Ethiopia, Mali, Malawi, Nigeria, Uganda, South Africa, Zambia and Zimbabwe).

    Sample size and sampling techniques

    A total of 302 580 pregnant women for stillbirth and abortion, 153 431 for birth weight and 115 556 births for preterm births were considered for final analysis. A multistage stratified systematic random sampling technique was employed to select study participants in the DHS. Since the aim of this was to assess the impact of SBI and LBI on adverse pregnancy outcomes, pregnant women who responded to the question preceding birth interval were eligible participants. The DHS data are available in public and open-access repository (https://dhsprogram.com/data/).35 For this study, we included the observations only of those who responded to the variable preceding birth interval, which means only mothers who had at least one previous birth were included.

    Study variables

    Outcome variables

    The dependent variable of this study was adverse birth outcomes, defined as ‘yes’ when a woman had at least one of the following: abortion, stillbirth, intrauterine growth restriction, congenital birth defects, gestational diabetes, gestational hypertensive disorder, LBW, macrosomia and preterm birth. In this study, given the availability of data in the DHS, we defined adverse pregnancy outcomes as if the pregnancy ends up with either of the following: abortion, stillbirth, LBW, macrosomia or preterm birth. Abortion among reproductive-aged women was derived from the DHS questions, ‘have you ever had a terminated pregnancy (v228)?’, ‘Century Month Code (CMC) pregnancy ended (v231)?’ and ‘gestational age at termination (v233)?’. Then, a woman has an abortion if she experienced termination of pregnancy before 28 weeks of gestation.36 Similarly, we derived stillbirth from the same question, defined as pregnancy losses occurring after 7 completed months of gestation. The third outcome variable was preterm birth, defined as births among reproductive-aged women before 37 weeks of gestation. We derived the variable preterm birth from the DHS question, ‘duration of pregnancy for the most recent birth (b20)’, defined as ‘yes’ if born before 37 completed weeks of gestation but after 28 weeks of gestation and ‘no’ if the birth was given after 9 or more completed months of gestational age.37 The fourth outcome variable was LBW defined as weight at birth less than 2500 g38; whereas macrosomia was defined as a birth weight of 4000 g, derived from the DHS question ‘birth weight in kilograms (m19)?’. In the DHS, birth weight data were collected from mothers who gave birth within 5 years before the survey either by accessing birth weight through record review or by the mother’s report by recalling the measured weight of the child at birth.

    Treatment variable

    The treatment variable was the preceding birth interval. Primigravida mothers were excluded from this study because of the treatment variable (SBI or LBI). We define a birth interval as short, optimal or long. The DHS collected information on the duration of preceding birth intervals for the most recent live births during the 5 years preceding the survey. Then, the duration of the birth interval was categorised as short, optimal and long. An SBI is defined as the WHO recommendation of fewer than 33 months, 33–59 months was referred to as the OBI and duration ≥60 months was categorised as an LBI. The reasoning is that both SBI and LBI increase the risk of adverse pregnancy outcomes. We carried out the analysis in two separate models (mlogit using OBI as the reference category, short vs optimal and long vs optimal): in the first model, birth interval was classified as SBI or OBI, and in the second model, the birth interval was compared between the LBI and OBI. The treatment group was those mothers who had an SBI or LBI, and the control group is those who had optimal birth spacing.

    Confounding variables

    Many maternal pre-intervention characteristics have been included in the model as they ensure a better chance that the PSM assumption holds. Variables that have an effect on SBIs or LBIs and adverse pregnancy outcomes at the same time but which are not affected by the treatment (SBI or LBI) were included. We used the direct acyclic graph (DAG) to present the assumed relationship between variables graphically to identify variables to adjust for confounding and other variables using DAGitty V.3 software (figures 1 and 2).39 DAGs allow the identification of the variable set or set sufficient to adjust for confounding and other biases, based on the variable relationships. Variables such as residence, country, maternal education, household wealth status, sex of household head, marital status, maternal working status, media exposure, the number of births before the current birth and maternal age of the preceding child (calculated as maternal age (v012) minus the age of the current child and preceding birth interval) were considered for matching. Confounding variables that have a significant association with the treatment and outcome variables were considered for matching.

    Figure 1
    Figure 1

    Direct acyclic graph used to select controlling variables. HH, household head; LBW, low birth weight; SBI, short birth interval.

    Figure 2
    Figure 2

    Direct acyclic graph used to select controlling variables. HH, household head; LBI, low birth interval; LBW, low birth weight.

    Data management and analysis

    All the reported results were based on the weighted data adjusted for design and non-response using the weighting variable (v005), strata (v021) and primary sampling unit (v023). In experimental studies, specifically randomised controlled trials (RCTs), the study participants are assigned to either the treatment or control group through randomisation. Randomisation can control both known and unknown confounders. Unlike RCTs, randomisation cannot be applied in observational studies to allocate study participants in either of the groups and therefore inherent imbalance of the observed variables introduces bias and influences the causal effect of the exposure. In situations where confounding variables can be measured, we can adjust for and treat the imbalance between groups. A function of the observed covariates, the so-called balancing score, can be used to correct the imbalance between the control and treatment groups. Based on the balancing score, the observed variables should be independent of the assignment of the exposure, that is, SBI, LBI or optimal birth spacing.

    The propensity score method is commonly used to balance the inequality of confounding variables in observational studies. Using PSM, the difference in outcomes between the child born to mothers with SBIs or LBIs and a child with OBIs will be an unbiased estimate of the effect of SBIs or LBIs after controlling for the observed covariates through propensity scores. A propensity score is the likelihood that a patient received a treatment (SBI or LBI) given all the observed covariates. It is a conditional probability of receiving treatment (SBI or LBI) and thus always has a value between 0 and 1. The larger the propensity score, the more likely a woman was to have SBI or LBI. The treatment variable of interest must be dichotomous in a propensity score analysis.

    Propensity score analysis usually starts with an assessment of the imbalance of the baseline covariates between the treatment and control groups. This can be assessed by significance tests like an independent t-test for continuous variables and Χ2 for categorical variables.

    In the propensity score model, the exposure variable is considered a dependent variable, whereas the observed confounding variables are considered predictors. Based on the relationships between treatment and outcome, observed covariates can be categorised into three groups: covariates only related to treatment assignment; covariates related to both treatment assignment and outcome (ie, confounders); and covariates only related to outcome. However, in this study, only confounders were included in the propensity score model, and to leave the covariates related only to the outcome. There are different propensity score adjustment methods such as covariate adjustment, inverse probability of treatment weight (IPTW), stratification and PSM.40–42 The most commonly used approach for controlling confounding by indications is PSM in non-experimental or observational studies. PSM and IPTW estimates are based on the propensity scores however have different interpretations, which may be more or less suitable under different conditions. Both of the methods wish to balance distributions of the study participants’ characteristics at baseline. PSM offers a more transparent method, which is readily understood and easy to communicate, and robust to misspecification of the propensity score compared with IPTW where extreme weights can bias the estimation of the treatment effect. On the other hand, IPTW is advantageous in retaining all study subjects in the analysis specifically, which is preferred when there are limitations in terms of sample size (when we have a small sample size).

    For our study, we preferred PSM because it minimises a greater proportion of the systematic differences in baseline characteristics between the treatment and control groups than stratification, covariate adjustment or IPTW in addition to the above-mentioned advantages. The likelihood of the pregnant women having SBI or LBI based on the selected confounders is reduced to a propensity score for each pregnant woman. This propensity score is generated for each subject from the selected confounders. Since the treatment variable of interest is dichotomous (SBI vs OBI and LBI vs OBI), the common methods adopted to produce propensity scores are either logistic regression or discriminant analysis.

    Embedded Image

    Where p(x) is the predicted probability of the event which is coded with ‘1’ SBI/LBI otherwise ‘0’; the same is true for adverse pregnancy outcomes. Then, to assess the impact of SBI and LBI on adverse pregnancy outcomes, estimation was done using PSM. This is the most popular statistical approach that can address the primary drawback of causal inference from observational research designs when randomisation cannot be used to establish the treatment and control groups. It forms matching sets of control and treatments of individuals whose propensity scores are similar. After a matching sample has been established, the impact of SBI or LBI can be assessed by comparing adverse pregnancy outcomes directly between SBI and OBI women, and LBI and OBI women in the matched sample.

    The PSM approach was chosen because the birth interval was not randomly assigned and can be affected strongly by observable and non-observable variables. Variables that have a significant association with birth interval and adverse pregnancy outcomes were selected as matching variables. Births with SBIs or LBIs were matched to births with OBIs using logit regression (psmatch2 STATA command). Besides, we used pstest to assess the balance for all covariates before and after matching, with a 5% level of significance or more considered indicative of imbalance.

    We aim to estimate the average effect of SBIs and LBIs on the treated. Assume AiT to be adverse pregnancy outcomes for those ith birth with SBIs or LBIs (treatment group), and AiC denotes adverse pregnancy outcomes for mothers with OBIs. The observed outcome can be written as Ai=(1−Ti) AiC+TiAiJ, where Ti=0, and 1 denotes treatment assignment (birth interval). The gain from the treatment (SBI) is (AiT−AiC) and our interest is to estimate the average treatment effect (ATE) (SBI or LBI) on the treated (ATT), E (AiT−AiC/Ti=1).43 This cannot be estimated directly since neither is normally observed as AiT for Ti=0 and AiC for Ti=1 are not known.

    Overall variables for matching were selected only if they are conceptualised before the treatment (SBI or LBI). The most common assumptions of PSM such as common support and selection of unobservable were assessed graphically and statistically. During analysis, the common support option was considered to limit testing of the balancing propensity to only mothers with treatment (SBI or LBI) whose propensity score for adverse pregnancy outcomes lay within ranges of propensity scores for controls. Using the pstest command to assess the covariate balance, we tested the following matching methods: nearest neighbour matching with and without replacement, and radius matching with callipers from 0.01. The psmatch2 command was used to generate the ATT and ATE for the matching method that produced the highest quality of matches. The common support option was also employed to produce higher-quality matches.

    The quality of matching was evaluated based on the balancing of the covariates between the treated and control groups. First, the quality of matching was assessed by computing the standardised bias before matching and after matching. The bias is computed as the percentage difference of the sample means in the treated and matched control groups as a percentage of the square root of the average of the sample variances in both groups. Though no hard and fast rule exists on the level of standardised difference that would indicate an imbalance, a difference of less than 10% is taken to indicate a negligible difference. Second, the pseudo R2 and likelihood-ratio test of the joint insignificance of all the covariates from the logit estimation of the conditional treatment probability before and after matching were used.

    A sensitivity analysis was conducted to test the robustness of the PSM estimates.44 As the outcome variables were binary, the Mantel-Haenszel test statistic was used to check whether the PSM estimates are sensitive to the hidden bias.45 The gamma coefficient is the factor by which the unobserved confounder or hidden bias affects the assignment of the intervention to the treated and control group. The gamma value ranges from 1 to 2 with a 0.05 increment using the mhbounds STATA command.46

    Patient and public involvement

    Patients and the public were not involved in this study since we have conducted a secondary data analysis based on already available DHS data, which were collected to provide estimates of common health and health-related indicators.

    Results

    Characteristics of the study participants

    A total of 302 580 women who gave birth within 5 years preceding the survey and had at least one previous birth were included, ranging from 1541 (0.51%) in Sao Tome to 27 382 (9.05%) in Nigeria (figure 3). About 141 748 (46.85%) had an SBI, 119 649 (39.54%) had an OBI and 41 183 (13.61%) had an LBI. The prevalence of LBW and macrosomia in SSA was 9.63% and 9.04%, respectively. The prevalence of stillbirth and preterm birth in SSA was 0.84% (95% CI: 0.81%, 0.87%) and 4.87% (95% CI: 4.75%, 4.99%), respectively. Of the total 302 580 pregnancies, 18 493 (6.11%) ended up with abortion (table 1). All the baseline characteristics such as country, residence, maternal education level, maternal age, maternal marital status, media exposure, maternal occupation status, sex of household head, parity and household wealth status had a significant association with birth interval, birth weight, abortion, stillbirth and preterm births (tables 2–4 and online supplemental table 1).

    Figure 3
    Figure 3

    The percentage of study participants across countries in sub-Saharan Africa.

    View this table:
    Table 1

    Characteristics of the study participants in 12 East African countries

    View this table:
    Table 2

    Association between baseline characteristics of the study participants with birth interval and birth weight in SSA

    View this table:
    Table 3

    Association between baseline characteristics of the study participants with birth interval and preterm births in SSA

    View this table:
    Table 4

    Logit regression analysis of factors associated with birth intervals in SSA

    Estimations of propensity score

    The logit model was for estimating the propensity score of SBI or LBI intervention in the study population. The strength of the association, the direction of the association and the significance of the estimates were in line with previous research findings (table 5). The mean propensity score was 0.5347 with minimal variability (SD=0.0001) between the intervention (SBI) and control (OBI) groups, and 0.2321 with minimal variability (SD=0.0002) between the intervention (LBI) and control (OBI) groups. We assessed the common support assumptions graphically and statistically, and the assumption fits (figures 4–13). Observations in the intervention and control groups with propensity score outside the region of common support were not included in the analysis.

    Figure 4
    Figure 4

    Histogram to show common support.

    Figure 5
    Figure 5

    Histogram to show common support.

    Figure 6
    Figure 6

    Histogram to show common support.

    Figure 7
    Figure 7

    Histogram to show common support.

    Figure 8
    Figure 8

    Histogram to show common support.

    Figure 9
    Figure 9

    Histogram to show common support.

    Figure 10
    Figure 10

    Histogram to show common support.

    Figure 11
    Figure 11

    Histogram to show common support.

    Figure 12
    Figure 12

    Histogram to show common support.

    Figure 13
    Figure 13

    Histogram to show common support.

    View this table:
    Table 5

    A propensity score-matched analysis of the impact of birth intervals on adverse pregnancy outcomes

    Impact of SBI and LBI on adverse pregnancy outcomes

    We estimated the impact of SBI and LBI on adverse pregnancy outcomes by the estimated difference in adverse birth outcomes between the treated groups (SBI or LBI) and the matched control groups (OBI). The PSM analysis estimates the impact of the treatment (SBI or LBI) by controlling the background variables which have an association with adverse birth outcomes and birth intervals. A radius matching with a calliper width of 0.01 had the best quality of matching and it was used to estimate the ATE of SBI and LBI among the population, ATT and average effect among the untreated. The unmatched estimate showed that births from pregnant women who had short preceding birth intervals had a 1.1%, 0.2%, 0.1%, and 2.6% increased risk of experiencing abortion, stillbirth, LBW, and macrosomia, respectively, whereas there is no risk difference in preterm births between pregnancy with OBI and SBI. The ATEs of SBI on abortion, stillbirth, LBW and preterm births were 0.5%, 0.1%, 0.2%, and 0.4%, respectively. This slightly higher effect of SBIs is due to the higher effect of SBI on the control groups (OBI), the effect when the pregnant women who had OBI could have SBIs. On the other hand, the ATE of SBIs on macrosomia was −1.4%, showing that pregnancies with SBI have led to a reduction of macrosomia by 1.4%. The ATT was 1%, 1%, and 1.1% increased risk of abortion, LBW, and preterm births, respectively, while there was no difference in risk of stillbirth between the treated and control groups. On the other hand, the ATT was −2.6%, which showed that pregnancies with SBI led to a reduction in the risk of macrosomia by 2.6% among the treated groups.

    Likewise, the difference in estimated ATE (SBI) on untreated groups in the treated and control groups was 0.2%, 0.3%, 0.6%, and 0.4% for abortion, stillbirth, LBW, and preterm birth, respectively. On the other hand, the ATE on untreated groups was −0.1%, showing that the control groups had a reduction in risk of macrosomia by 0.1%.

    The ATEs of LBIs on abortion, stillbirth, LBW, macrosomia and preterm births were 1.3%, 0.4%, 1.0%, 3.4%, and 0.2%, respectively. The ATT had 0.9%, 0.4%, 2.4%, 2.8%, and 0.2% increased risk of abortion, stillbirth, LBW, macrosomia, and preterm births, respectively (table 5).

    Quality of matching

    Common support

    When we fitted the propensity-matched analysis of the impact of the SBI, about 14, 13, 5 and 8 observations were dropped due to the common support option for abortion, stillbirth, LBW and macrosomia, respectively, while no observation was dropped for preterm births. In addition, in the PSM analysis of the impact of LBIs on adverse birth outcomes, 13, 10, 7, 10 and 30 observations were discarded due to off support (table 6). We plot a histogram to visualise the distributions of propensity scores and the distribution is almost similar for both the groups’ post-matching on propensity score (figures 4–13). The presence of significant overlap between the characteristics of the treated and control groups proves the validity of the common support assumption.

    View this table:
    Table 6

    Common support

    Sensitivity analysis

    We used the Rosenbaum bounding approach to determine how strongly unmeasured variables (hidden bias) influence the selection process to determine the implications of the matching analysis. In all of the analyses, in a study free of bias, that is, where Ґ=1, the QMH statistics is… and would constitute strong evidence that SBI and LBI cause adverse pregnancy outcomes. The upper bound on the significance level for Ґ=1.05, 1.1, 1.15……2 was significant and showed that the study is insensitive to hidden bias (tables 7–16).

    View this table:
    Table 7

    Quality of matching for the impact of SBI on abortion

    View this table:
    Table 8

    Quality of matching for the impact of LBI on abortion

    View this table:
    Table 9

    Quality of matching for the impact of SBI on stillbirth

    View this table:
    Table 10

    Quality of matching for the impact of LBI on stillbirth

    View this table:
    Table 11

    Quality of matching for the impact of SBI on LBW

    View this table:
    Table 12

    Quality of matching for the impact of LBI on LBW

    View this table:
    Table 13

    Quality of matching for the impact of SBI on macrosomia

    View this table:
    Table 14

    Quality of matching for the impact of LBI on macrosomia

    View this table:
    Table 15

    Quality of matching for the impact of SBI on preterm births

    View this table:
    Table 16

    Quality of matching for the impact of LBI on preterm births

    Discussion

    In public health, healthcare decision-makers attempted to assess how the public health interventions among the treated populations have changed in the absence of the programme. This study investigated the causal impact of SBI and LBI on adverse pregnancy outcomes using PSM analysis. This approach is one of the best approaches to assess the impact of a certain intervention in observational studies by constructing an adequate comparison group since randomisation is not possible. Previous researchers found SBI and LBI as significant predictors of adverse pregnancy outcomes and they have reported the association between SBI/LBI and adverse pregnancy outcomes in SSA countries. However, we estimated the actual impact of SBI/LBI on abortion, stillbirth, LBW, macrosomia and preterm births selection of unobservable would be an issue of concern. Pregnancies with SBI/LBI have socioeconomically, demographically and birth-related characteristics that are different at baseline from pregnancies with an OBI. Therefore, conducting a standard regression analysis of adverse pregnancy outcomes between pregnancies with SBI/LBI and pregnancies with OBI would yield biased estimates of the impact of SBI/LBI and the result would not be adjusted for endogeneity bias. This study found that the prevalence of SBI and LBI in SSA was 46.85% (95% CI: 46.67%, 47.02%) and 13.61% (95% CI: 13.49%, 13.73%), respectively. The higher prevalence of SBI in SSA might be due to the serious lack of healthcare workers and healthcare services’ inaccessibility to provide reproductive and maternal healthcare services in the majority of African countries.47 48

    In the PSM analysis, we found that ATEs of SBIs on abortion, stillbirth, LBW, and preterm births were 0.5%, 0.1%, 0.2%, and 0.4%, respectively. This slightly higher effect of SBIs is due to the higher effect of SBI on the control groups (OBI), the effect when the pregnant women who had OBI could have SBIs. On the other hand, the ATE of SBIs on macrosomia was −1.4%, showing that pregnancies with SBI have led to a reduction of macrosomia by 1.4%. The ATT was 1%, 1%, and 1.1% increased risk of abortion, LBW and preterm births, respectively, while there was no difference in risk of stillbirth between the treated and control groups. On the other hand, the ATT was −2.6%, which showed that pregnancies with SBI led to a reduction in the risk of macrosomia by 2.6% among the treated groups. This positive causal impact of SBI on abortion, stillbirth, LBW, macrosomia and preterm births can be attributed to the fact that SBIs cause maternal nutritional depletion, specifically folate depletion, which could lead to congenital anomaly and the pregnancy ending up with stillbirth or abortion.29 49 Mothers who had SBIs are prone to undernutrition because of inadequate time to recover from the physiological stresses of subsequent pregnancy.26 31 50 Furthermore, disease transmission could be common when the interbirth interval is too short.51 Moreover, SBIs prevent a mother from recuperating in their nutrition and physical status, which is responsible for poor fetal development like LBW, stillbirth, abortion and preterm birth.17 32 It is because micronutrients and muscle mass are being depleted from the previous pregnancy and need a minimum of 2 years to completely restore the nutritional and physical status.52 That is why the WHO recommends an interbirth interval of at least 2 years (24 months),34 which corresponds to a birth-to-birth interval of 33 months under the assumption of 9 months of pregnancy to reduce the incidence of adverse pregnancy outcomes.23

    The ATEs of LBIs on abortion, stillbirth, LBW, macrosomia and preterm births were 1.3%, 0.4%, 1.0%, 3.4%, and 0.2%, respectively. The ATT had 0.9%, 0.4%, 2.4%, 2.8%, and 0.2% increased risk of abortion, stillbirth, LBW, macrosomia, and preterm births, respectively. The increased risk of adverse pregnancy outcomes for pregnancies that had LBIs may be due to corresponding factors such as advanced maternal age, and the decline in the maternal physiological and anatomical capacities of the reproductive system.16 19 Besides, in low/middle-income countries, the LBI is a reflection of poor and inconsistent contraceptive uses as contraception failure is the most common factor for a prolonged birth interval; this could increase the risk of complications like pre-eclampsia, obstructed labour, uterine fibrosis, etc.27 53 54

    Even though this study offers important insight into the actual impact of SBI/LBI on adverse pregnancy outcomes, the result should be interpreted in light of the following limitations. The matching was done based on the observed variables only, and there is a possibility of residual confounding (unobserved variables). Moreover, the DHS is a cross-sectional study and it is prone to social desirability and recall bias. Despite the above-mentioned limitations, the study has the following strengths. First, this study is based on nationally representative DHS data with a high response rate. Second, the DHS uses a standardised questionnaire for the data collection, which is consistent across all 36 countries. Furthermore, this study is the adjustment for potential confounders using the PSM approach in the estimation of the association between SBI/LBI and adverse pregnancy outcomes.

    Conclusions

    SBI was a risk factor for stillbirth, abortion, LBW and preterm births while protective against macrosomia. Similarly, LBI was a risk factor for stillbirth, abortion, LBW, macrosomia and preterm births. These findings evidenced that public health programmes targeting improving maternal and child health should empower reproductive-aged women to have optimal birth spacing to reduce the incidence of stillbirth, abortion, LBW, macrosomia and preterm births in SSA. We further recommend a study that will examine the causal impact of SBI and LBI on adverse pregnancy outcomes by incorporating more proximal variables which were not observed in this study.

    Data availability statement

    Data are available in a public, open access repository. Data are available in a public, open access repository. These data can be found from https://dhsprogram.com/data/dataset_admin/login_main.cfm?CFID=52406619&CFTOKEN=acf80f39c3b63be9-65910752-FB64-58E6-E6F9496A4F66E431.

    Ethics statements

    Patient consent for publication

    Ethics approval

    Since the study was a secondary data analysis of publicly available survey data from the MEASURE DHS Programme, ethical approval and participant consent were not necessary for this particular study. We requested the DHS Programme, and permission was granted to download and use the data for this study from http://www.dhsprogram.com. There are no names of individuals or household addresses in the data files.

    References

    1. ↵
      1. Kramer MS
      . The epidemiology of adverse pregnancy outcomes: an overview. J Nutr 2003;133:1592S–1596S. doi:10.1093/jn/133.5.1592S
      OpenUrlAbstract/FREE Full Text
    2. ↵
      1. Gardosi J,
      2. Francis A
      . Adverse pregnancy outcome and association with small for gestational age birthweight by customized and population-based percentiles. Am J Obstet Gynecol 2009;201:28. doi:10.1016/j.ajog.2009.04.034
      OpenUrlPubMed
    3. ↵
      1. Kyei-Nimakoh M,
      2. Carolan-Olah M,
      3. McCann TV
      . Access barriers to obstetric care at health facilities in sub-saharan africa-a systematic review. Syst Rev 2017;6:110. doi:10.1186/s13643-017-0503-x
      OpenUrlCrossRefPubMed
    4. ↵
      1. Mombo-Ngoma G,
      2. Mackanga JR,
      3. González R, et al
      . Young adolescent girls are at high risk for adverse pregnancy outcomes in sub-Saharan Africa: an observational multicountry study. BMJ Open 2016;6:e011783. doi:10.1136/bmjopen-2016-011783
    5. ↵
      1. Dellicour S,
      2. Sevene E,
      3. McGready R, et al
      . First-Trimester artemisinin derivatives and quinine treatments and the risk of adverse pregnancy outcomes in Africa and Asia: a meta-analysis of observational studies. PLoS Med 2017;14:e1002290. doi:10.1371/journal.pmed.1002290
    6. ↵
      1. Kurth F,
      2. Bélard S,
      3. Mombo-Ngoma G, et al
      . Adolescence as risk factor for adverse pregnancy outcome in central africa--a cross-sectional study. PLoS One 2010;5:e14367. doi:10.1371/journal.pone.0014367
    7. ↵
      1. Lawn JE,
      2. Gravett MG,
      3. Nunes TM, et al
      . Global report on preterm birth and stillbirth (1 of 7): definitions, description of the burden and opportunities to improve data. BMC Pregnancy Childbirth 2010;10 Suppl 1:1–22. doi:10.1186/1471-2393-10-S1-S1
      OpenUrl
    8. ↵
      1. World Health Organization
      . Global nutrition targets 2025: low birth weight policy brief; 2014.
    9. ↵
      1. Henshaw SK,
      2. Singh S,
      3. Haas T
      . The incidence of abortion worldwide. Int Fam Plann Persp 1999;25:S30–8. doi:10.2307/2991869
      OpenUrlCrossRefPubMedWeb of Science
    10. ↵
      1. World Health Organization
      . Global nutrition targets 2025: low birth weight policy brief. 2014.
    11. ↵
      1. Ye J,
      2. Torloni MR,
      3. Ota E, et al
      . Searching for the definition of macrosomia through an outcome-based approach in low- and middle-income countries: a secondary analysis of the who global survey in Africa, Asia and Latin America. BMC Pregnancy Childbirth 2015;15:324. doi:10.1186/s12884-015-0765-z
      OpenUrl
    12. ↵
      1. Tela FG,
      2. Bezabih AM,
      3. Adhanu AK, et al
      . Fetal macrosomia and its associated factors among singleton live-births in private clinics in Mekelle city, Tigray, Ethiopia. BMC Pregnancy Childbirth 2019;19:219. doi:10.1186/s12884-019-2379-3
      OpenUrl
    13. ↵
      1. Reyes L,
      2. Mañalich R
      . Long-term consequences of low birth weight. Kidney Int Suppl 2005:S107–11. doi:10.1111/j.1523-1755.2005.09718.x
    14. ↵
      1. Pimentel J,
      2. Ansari U,
      3. Omer K, et al
      . Factors associated with short birth interval in low- and middle-income countries: a systematic review. BMC Pregnancy Childbirth 2020;20:156. doi:10.1186/s12884-020-2852-z
      OpenUrlPubMed
    15. ↵
      1. Shifti DM,
      2. Chojenta C,
      3. Holliday E, et al
      . Effects of short birth interval on neonatal, infant and under-five child mortality in Ethiopia: a nationally representative observational study using inverse probability of treatment weighting. BMJ Open 2021;11:e047892. doi:10.1136/bmjopen-2020-047892
    16. ↵
      1. Kozuki N,
      2. Walker N
      . Exploring the association between short/long preceding birth intervals and child mortality: using reference birth interval children of the same mother as comparison. BMC Public Health 2013;13 Suppl 3:1–10. doi:10.1186/1471-2458-13-S3-S6
      OpenUrl
    17. ↵
      1. Exavery A,
      2. Mrema S,
      3. Shamte A, et al
      . Levels and correlates of non-adherence to who recommended inter-birth intervals in Rufiji, Tanzania. BMC Pregnancy Childbirth 2012;12:1–8. doi:10.1186/1471-2393-12-152
      OpenUrlCrossRefPubMed
    18. ↵
      1. Molitoris J,
      2. Barclay K,
      3. Kolk M
      . When and where birth spacing matters for child survival: an international comparison using the DHS. Demography 2019;56:1349–70. doi:10.1007/s13524-019-00798-y
      OpenUrlPubMed
    19. ↵
      1. Nisha MK,
      2. Alam A,
      3. Islam MT, et al
      . Risk of adverse pregnancy outcomes associated with short and long birth intervals in Bangladesh: evidence from six Bangladesh demographic and health surveys, 1996-2014. BMJ Open 2019;9:e024392. doi:10.1136/bmjopen-2018-024392
    20. ↵
      1. Hanley GE,
      2. Hutcheon JA,
      3. Kinniburgh BA, et al
      . Interpregnancy interval and adverse pregnancy outcomes. Obstet Gynecol 2017;129:408–15. doi:10.1097/AOG.0000000000001891
      OpenUrlPubMed
    21. ↵
      1. Bener A,
      2. Saleh NM,
      3. Salameh KMK, et al
      . The impact of the interpregnancy interval on birth weight and other pregnancy outcomes. Rev Bras Saude Mater Infant 2012;12:233–41. doi:10.1590/S1519-38292012000300003
      OpenUrl
    22. ↵
      1. Mahande MJ,
      2. Obure J
      . Effect of interpregnancy interval on adverse pregnancy outcomes in Northern Tanzania: a registry-based retrospective cohort study. BMC Pregnancy Childbirth 2016;16:140. doi:10.1186/s12884-016-0929-5
      OpenUrlPubMed
    23. ↵
      1. World Health Organization
      . Report of a WHO technical consultation on birth spacing: geneva, switzerland 13-15 June 2005. 2007.
    24. ↵
      1. DaVanzo J,
      2. Hale L,
      3. Razzaque A, et al
      . Effects of interpregnancy interval and outcome of the preceding pregnancy on pregnancy outcomes in Matlab, Bangladesh. BJOG 2007;114:1079–87. doi:10.1111/j.1471-0528.2007.01338.x
      OpenUrlCrossRefPubMedWeb of Science
    25. ↵
      1. Hegelund ER,
      2. Urhoj SK,
      3. Andersen A-MN, et al
      . Interpregnancy interval and risk of adverse pregnancy outcomes: a register-based study of 328,577 pregnancies in Denmark 1994–2010. Matern Child Health J 2018;22:1008–15. doi:10.1007/s10995-018-2480-7
      OpenUrl
    26. ↵
      1. Conde-Agudelo A,
      2. Rosas-Bermudez A,
      3. Castaño F, et al
      . Effects of birth spacing on maternal, perinatal, infant, and child health: a systematic review of causal mechanisms. Stud Fam Plann 2012;43:93–114. doi:10.1111/j.1728-4465.2012.00308.x
      OpenUrlCrossRefPubMed
    27. ↵
      1. Kozuki N,
      2. Lee ACC,
      3. Silveira MF, et al
      . The associations of birth intervals with small-for-gestational-age, preterm, and neonatal and infant mortality: a meta-analysis. BMC Public Health 2013;13 Suppl 3:1–9. doi:10.1186/1471-2458-13-S3-S3
      OpenUrl
    28. ↵
      1. Smith GCS
      . Interpregnancy interval and risk of preterm birth and neonatal death: retrospective cohort study. BMJ 2003;327:313–0. doi:10.1136/bmj.327.7410.313
      OpenUrlAbstract/FREE Full Text
    29. ↵
      1. Winkvist A,
      2. Rasmussen KM,
      3. Habicht J-P
      . A new definition of maternal depletion syndrome. Am J Public Health 1992;82:691–4. doi:10.2105/ajph.82.5.691
      OpenUrlCrossRefPubMedWeb of Science
    30. ↵
      1. Dhingra S,
      2. Pingali PL
      . Effects of short birth spacing on birth-order differences in child stunting: evidence from India. Proc Natl Acad Sci USA 2021;118. doi:10.1073/pnas.2017834118
    31. ↵
      1. King JC
      . The risk of maternal nutritional depletion and poor outcomes increases in early or closely spaced pregnancies. J Nutr 2003;133:1732S–1736S. doi:10.1093/jn/133.5.1732S
      OpenUrlAbstract/FREE Full Text
    32. ↵
      1. DaVanzo J,
      2. Hale L,
      3. Razzaque A, et al
      . The effects of pregnancy spacing on infant and child mortality in Matlab, Bangladesh: how they vary by the type of pregnancy outcome that began the interval. Popul Stud (Camb) 2008;62:131–54. doi:10.1080/00324720802022089
      OpenUrlCrossRefPubMed
    33. ↵
      1. Barros FC,
      2. Bhutta ZA,
      3. Batra M, et al
      . Global report on preterm birth and stillbirth (3 of 7): evidence for effectiveness of interventions. BMC Pregnancy Childbirth 2010;10 Suppl 1:1–36. doi:10.1186/1471-2393-10-S1-S3
      OpenUrl
    34. ↵
      1. Rutstein SO
      . Effects of preceding birth intervals on neonatal, infant and under-five years mortality and nutritional status in developing countries: evidence from the demographic and health surveys. Int J Gynaecol Obstet 2005;89 Suppl 1:S7–24. doi:10.1016/j.ijgo.2004.11.012
      OpenUrlCrossRefPubMed
    35. ↵
      1. Macro I
      . Measure DHS. 2011. Available: http://www measuredhs com
    36. ↵
      1. Lim LM,
      2. Singh K
      . Termination of pregnancy and unsafe abortion. Best Pract Res Clin Obstet Gynaecol 2014;28:859–69. doi:10.1016/j.bpobgyn.2014.05.005
      OpenUrl
    37. ↵
      1. Vogel JP,
      2. Chawanpaiboon S,
      3. Moller A-B, et al
      . The global epidemiology of preterm birth. Best Pract Res Clin Obstet Gynaecol 2018;52:3–12. doi:10.1016/j.bpobgyn.2018.04.003
      OpenUrlCrossRefPubMed
    38. ↵
      1. World Health Organization
      . International statistical classification of diseases and related health problems: alphabetical index, 3. 2004.
    39. ↵
      1. Textor J,
      2. Hardt J,
      3. Knüppel S
      . DAGitty: a graphical tool for analyzing causal diagrams. Epidemiology 2011;22:745. doi:10.1097/EDE.0b013e318225c2be
      OpenUrlPubMed
    40. ↵
      1. Deb S,
      2. Austin PC,
      3. Tu JV, et al
      . A review of propensity-score methods and their use in cardiovascular research. Can J Cardiol 2016;32:259–65. doi:10.1016/j.cjca.2015.05.015
      OpenUrlCrossRefPubMed
    41. ↵
      1. Lanehart RE,
      2. Rodriguez de Gil P,
      3. Kim ES, et al
      . Propensity score analysis and assessment of propensity score approaches using SAS procedures. In: Proceedings of the SAS Global Forum 2012 Conference: 2012. Cary North Carolina: SAS Institute Inc, 2012: 22–5.
    42. ↵
      1. Stone CA,
      2. Tang Y
      . Comparing propensity score methods in balancing covariates and recovering impact in small sample educational program evaluations. Pract Assess Res Eval 2013;18:13. doi:10.7275/qkqa-9k50
    43. ↵
      1. Baser O
      . Too much ado about propensity score models? Comparing methods of propensity score matching. Value Health 2006;9:377–85. doi:10.1111/j.1524-4733.2006.00130.x
      OpenUrlCrossRefPubMedWeb of Science
    44. ↵
      1. Becker SO,
      2. Caliendo M
      . Sensitivity analysis for average treatment effects. Stata J 2007;7:71–83. doi:10.1177/1536867X0700700104
      OpenUrlWeb of Science
    45. ↵
      1. Li L,
      2. Shen C,
      3. Wu AC, et al
      . Propensity score-based sensitivity analysis method for uncontrolled confounding. Am J Epidemiol 2011;174:345–53. doi:10.1093/aje/kwr096
      OpenUrlCrossRefPubMed
    46. ↵
      1. Dehejia R
      . Practical propensity score matching: a reply to smith and todd. J Econom 2005;125:355–64. doi:10.1016/j.jeconom.2004.04.012
      OpenUrlCrossRefWeb of Science
    47. ↵
      1. Haider TL,
      2. Sharma M
      . Barriers to family planning and contraception uptake in sub-Saharan Africa: a systematic review. Int Q Community Health Educ 2012;33:403–13. doi:10.2190/IQ.33.4.g
      OpenUrlPubMed
    48. ↵
      1. Lucas D
      . Fertility and family planning in southern and central Africa. Stud Fam Plann 1992;23:145–58. doi:10.2307/1966724
      OpenUrlCrossRefPubMedWeb of Science
    49. ↵
      1. Miller JE,
      2. Huss-Ashmore R
      . Do reproductive patterns affect maternal nutritional status?: An analysis of maternal depletion in lesotho. Am J Hum Biol 1989;1:409–19. doi:10.1002/ajhb.1310010403
      OpenUrlCrossRef
    50. ↵
      1. Rousso D,
      2. Panidis D,
      3. Gkoutzioulis F, et al
      . Effect of the interval between pregnancies on the health of mother and child. Eur J Obstet Gynecol Reprod Biol 2002;105:4–6. doi:10.1016/s0301-2115(02)00077-5
      OpenUrlCrossRefPubMedWeb of Science
    51. ↵
      1. Boerma JT,
      2. Bicego GT
      . Preceding birth intervals and child survival: searching for pathways of influence. Stud Fam Plann 1992;23:243–56. doi:10.2307/1966886
      OpenUrlCrossRefPubMed
    52. ↵
      1. Kudeva R,
      2. Halaas B,
      3. Kagotho N, et al
      . Optimal birth interval and empowerment: a closer look at women’s agency in kenya. JMWH 2020;14:1–10. doi:10.12968/ajmw.2019.0028
      OpenUrl
    53. ↵
      1. Zhang J
      . Partner change, birth interval and risk of pre-eclampsia: a paradoxical triangle. Paediatr Perinat Epidemiol 2007;21 Suppl 1:31–5. doi:10.1111/j.1365-3016.2007.00835.x
      OpenUrl
    54. ↵
      1. Moultrie TA,
      2. Sayi TS,
      3. Timæus IM
      . Birth intervals, postponement, and fertility decline in Africa: a new type of transition? Popul Stud (Camb) 2012;66:241–58. doi:10.1080/00324728.2012.701660
      OpenUrlPubMed

    Supplementary materials

    • Supplementary Data

      This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Footnotes

    • Contributors Garantor- GAT, Conceptualisation—GAT, ABT, YY, DA and ALM. Data curation—GAT, ABT, YY, DA and ALM. Funding acquisition—GAT, ABT, YY, DA and ALM. Investigation—GAT, ABT, YY, DA and ALM. Methodology—GAT, ABT, YY, DA and ALM. Project administration—GAT, ABT, YY, DA and ALM. Resources—GAT, ABT, YY, DA and ALM. Software—GAT, ABT, YY, DA and ALM. Supervision—GAT, ABT, YY, DA and ALM. Validation—GAT, ABT, YY, DA and ALM. Visualisation—GAT, ABT, YY, DA and ALM. Writing—GAT, ABT, YY, DA and ALM. Review and editing—GAT, ABT, YY, DA and ALM. All the authors read and approved the manuscript.

    • Funding The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

    • Competing interests None declared.

    • Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting, or dissemination plans of this research.

    • Provenance and peer review Not commissioned; externally peer reviewed.

    • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.