You are here

Article Text

Article menu
Global health
Original research
Validation of the Pediatric Resuscitation and Trauma Outcome (PRESTO) model in injury patients in Tanzania
  1. Elizabeth M Keating1,
  2. Modesta Mitao2,3,
  3. Arthi Kozhumam4,
  4. Joao Vitor Souza5,
  5. Cecilia S Anthony2,3,
  6. Dalton Breno Costa6,
  7. Catherine A Staton7,8,
  8. Blandina T Mmbaga2,3,9,
  9. Joao Ricardo Nickenig Vissoci7,8
  1. 1Department of Pediatrics, Division of Pediatric Emergency Medicine, University of Utah, Salt Lake City, Utah, USA
  2. 2Kilimanjaro Christian Medical Centre, Moshi, Tanzania, United Republic of
  3. 3Kilimanjaro Christian Medical University College, Moshi, Tanzania, United Republic of
  4. 4Feinberg School of Medicine, Northwestern University, Chicago, Illinois, USA
  5. 5State University of Maringa, Maringa, Paraná, Brazil
  6. 6Department of Computer Science, University of North Carolina at Greensboro (UNCG), Greensboro, North Carolina, USA
  7. 7Department of Emergency Medicine, Duke University Medical Center, Durham, North Carolina, USA
  8. 8Duke Global Health Institute, Duke University, Durham, North Carolina, USA
  9. 9Kilimanjaro Clinical Research Institute, Moshi, Tanzania, United Republic of
  1. Correspondence to Dr Elizabeth M Keating; elizabeth.keating{at}hsc.utah.edu

Abstract

Introduction Sub-Saharan Africa has the highest rate of unintentional paediatric injury deaths. The Pediatric Resuscitation and Trauma Outcome (PRESTO) model predicts mortality using patient variables available in low-resource settings: age, systolic blood pressure (SBP), heart rate (HR), oxygen saturation, need for supplemental oxygen (SO) and neurologic status (Alert Verbal Painful Unresponsive (AVPU)). We sought to validate and assess the prognostic performance of PRESTO for paediatric injury patients at a tertiary referral hospital in Northern Tanzania.

Methods This is a cross-sectional study from a prospective trauma registry from November 2020 to April 2022. We performed exploratory analysis of sociodemographic variables and developed a logistic regression model to predict mortality using R (V.4.1). The logistic regression model was evaluated using area under the receiver operating curve (AUC).

Results 499 patients were enrolled with a median age of 7 years (IQR 3.41–11.18). 65% were boys, and in-hospital mortality was 7.1%. Most were classified as alert on AVPU Scale (n=326, 86%) and had normal SBP (n=351, 98%). Median HR was 107 (IQR 88.5–124). The logistic regression model based on the original PRESTO model revealed that AVPU, HR and SO were statistically significant to predict in-hospital mortality. The model fit to our population revealed AUC=0.81, sensitivity=0.71 and specificity=0.79.

Conclusion This is the first validation of a model to predict mortality for paediatric injury patients in Tanzania. Despite the low number of participants, our results show good predictive potential. Further research with a larger injury population should be done to improve the model for our population, such as through calibration.

  • ACCIDENT & EMERGENCY MEDICINE
  • Paediatric A&E and ambulatory care
  • Paediatric intensive & critical care
  • PAEDIATRIC SURGERY
  • TRAUMA MANAGEMENT

Data availability statement

The data for this manuscript is covered under a data sharing agreement, and thus we are unable to openly share these data. For data access requests, please contact our non-author KCMC representative Gwamaka William at gwamakawilliam14{at}gmail.com.

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-2022-070747

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

    • This is a cross-sectional study of 499 paediatric patients from a prospective paediatric trauma registry in Northern Tanzania from November 2020 to April 2022.

    • A multivariable logistic regression model was built to predict in-hospital mortality.

    • This study has a relatively limited sample size with 33 deaths.

    • We performed multiple imputation methods to deal with the 67 children with missing data and thus feel that our conclusions are sound.

    Introduction

    Each day, 1900 children and adolescents die from an injury,1 with the vast majority of these being unintentional2 and occurring in low-income and middle-income countries (LMICs).3 Injuries are the third leading cause of death in Sub-Saharan Africa and a cause of significant childhood disability.2 In Tanzania alone, one in five households have reported child and adolescent injuries.2 Despite this burden, a recent systematic review found that less than 15% of existing childhood injury-focused studies occur in Sub-Saharan Africa, and represent only 3 of the 48 countries in Sub-Saharan Africa.4

    Existing studies on paediatric trauma provide incomplete data and rely on hospital-based health records that do not reflect risk and protective factors. Further, they may underestimate the burden of paediatric injuries.2 In addition, the most common mechanisms of injury and outcomes of childhood injury have been shown to be country and region dependent.4 Most trauma predictor scores developed and used in high-income countries (HICs) require variables not commonly available in LMIC settings.5 In order to address the need for LMIC-specific paediatric trauma outcome indicators, St-Louis et al developed the Pediatric Resuscitation and Trauma Outcome (PRESTO) model for predicting short-term mortality using initial patient assessment variables appropriate for and available in low-resource settings.5 The PRESTO model has been shown to outperform other trauma scores in children less than 5 years of age.6

    The PRESTO model requires only patient’s age, noted hypotension via systolic blood pressure (SBP), heart rate (HR), oxygen saturation, need for supplemental oxygen (SO) and neurologic status (measured via the Alert Verbal Painful Unresponsive (AVPU) Scale). It was initially validated using the United States National Trauma Data Bank and has subsequently been validated in a middle-income setting (South Africa) and an LMIC setting (Rwanda).5–7 In its application and validation in Rwanda, it was shown to be similarly effective in predicting child mortality as the Revised Trauma Score and Kampala Trauma Score, and to outperform the Kampala Trauma Score in children less than 5 years of age.6

    Due to PRESTO’s success in both low-income6 and middle-income7 settings, and potential differences in causes, presentations and outcomes of specific childhood injuries in Tanzania, we sought to validate this scale in our population of paediatric injury patients at a tertiary zonal referral hospital in Northern Tanzania. We also sought to assess the effectiveness of the PRESTO model in prognostic performance of injury-related mortality in our population.

    Methods

    Study setting

    This study was conducted at Kilimanjaro Christian Medical Center (KCMC) in Moshi, Tanzania. KCMC is located in a semiurban area and is a tertiary zonal referral hospital in Kilimanjaro that serves the Northern zone of Tanzania and covers about 15 million people from Arusha, Kilimanjaro, Manyara and Tanga. The emergency medicine department at KCMC attends approximately 1400–1700 paediatric patients per year.

    Study design and population

    This is a cross-sectional study of 499 paediatric patients from a prospective paediatric trauma registry in Northern Tanzania from November 2020 to April 2022. The paediatric trauma registry was established at KCMC in November 2020. It is ongoing with prospective consecutive enrolment of all patients less than 18 years of age who present to KCMC emergency department for management of an injury that occurred within 1 month of presentation. Further details on the paediatric trauma registry data collection are available in Keating et al.8 Potential bias was addressed by enrolling all paediatric injury patients who presented regardless of age, sex or race. Patients with missing outcome variables or who died prior to arrival were excluded.

    Patient and public involvement

    There was no patient or public involvement in this paper.

    Data collection and study variables

    Data was collected by trained research assistants and input into REDCap.9 10 The dependent variable was in-hospital mortality. The independent variables identified for inclusion into the study analysis were patient demographics, mechanism of injury, SBP, HR, oxygen saturation, need for SO, neurologic status classified using the AVPU) Scale11 and in-hospital mortality. Variables used to calculate the PRESTO Score are defined in table 1.

    View this table:
    Table 1

    Variables to calculate the Pediatric Resuscitation and Trauma Outcome (PRESTO) model

    Data analysis

    Data analysis was performed in R Software for Statistical Computing (V.4.1).12 Descriptive statistics, such as means, dispersions and frequencies, were determined for sociodemographic characteristics. A multivariable logistic regression model was built to predict in-hospital mortality. A p-value of<0.05 was considered statistically significant.

    The variables included in the logistic regression model were selected according to the original PRESTO model,5 namely: age, noted hypotension (SBP), HR, O2 saturation, need for SO (airway support) and neurologic status (AVPU). We also tested if delays to presentation to definitive care at KCMC significantly impacted our model performance. The difference between date/time of injury and date/time of arrival to KCMC was measured in days and hours and was included in the model as a numeric variable (online supplemental tables S1 and S2).

    Multiple imputation methods using chain equations were used to handle missing data, using the Multiple Imputation by Chained Equations package.13 Briefly, data was treated as missing at random and 15 imputed versions of the dataset were created. Afterwards, pooling of model coefficients was performed and sensitivity analysis was done. The logistic regression model was evaluated using the area under the receiver operating characteristic curve (AUC), with specificity and sensitivity used to select the decision rule threshold (t) value. The receiver operating characteristic curve was constructed for different models whose AUC were compared to obtain best models. Data was randomly split into training and validation sets in a proportion of 75% and 25%, respectively. Stratification was performed to ensure that the proportion of cases and non-cases would be the same in both sets. The model was fit to the training data and a final evaluation was performed on the validation set, which behaves as ‘never seen before’ data.

    Finally, coefficients and thresholds obtained in previous publications5 6 were applied to our sample to test whether models built in populations with different characteristics (sample size, region, ethnicity, etc) would be able to show improved performance.

    Results

    A total of 499 participants were enrolled in the registry with a median age of 7 years old (IQR 4.00–11.1, unimodal mode of 6). Most participants were boys (n=316, 63%) and the most common mechanism of injury was road traffic injury (n=178, 36%). In-hospital mortality was recorded for 6.7% (n=33) of participants. Most (n=415, 86%) were classified as alert on the AVPU Scale, had normal SBP (n=451, 97%) and had a median HR of 106 (IQR 89–124). A total of 52 (10%) patients needed SO via nasal cannula, nasopharyngeal airway or intubation (table 2). The median time between injury and arrival to KCMC was 7 hours (IQR 3–33).

    View this table:
    Table 2

    Descriptive analysis of study sample and comparisons between groups with and without in-hospital mortality

    Neurologic status (AVPU), hypotension, HR, SO, oxygen saturation and SBP were statistically different (p<0.05) among patients who died versus those who survived (table 2). Stratification by age revealed that the distribution of HR and SBP did not differ among patients who died or survived, except for HR being statistically higher for teens who died (figure 1) and SBP being statistically lower for children in preschool age.

    Figure 1
    Figure 1

    Distribution of systolic blood pressure (A) and heart rate (B) according to age and in-hospital mortality. *p<0.05 as per Wilcoxon rank sum test.

    Univariate logistic regression models fit to the study population revealed that neurologic status (AVPU), hypotension, HR, SO and oxygen saturation were statistically significant (p<0.05) to predict in-hospital mortality (table 3).

    View this table:
    Table 3

    Crude ORs and CIs for univariate complete case and imputed multivariable logistic regressions considering Pediatric Resuscitation and Trauma Outcome (PRESTO) variables using the Tanzania dataset

    Among the 499 participants, 432 (86.6%) had complete information for the variables studied. Multiple imputations were applied to impute missing data and 15 versions of the dataset were generated. Multivariable logistic regression models were fit to the 15 versions and the pooled ORs and confidence intervals are displayed in table 3. The logistic regression model executed following the original PRESTO model5 revealed that abnormal neurologic status (AVPU) and SO were statistically significant (p<0.05) to predict in-hospital mortality (table 3). Statistically significant differences were found for participants who were either responding to verbal stimuli (4.45, CI 1.29 to 15.8, p=0.019) or unresponsive (5.70, CI 1.08 to 30.1, p=0.040), compared with those who were alert.

    For each model, we set the best threshold (t) value that maximised sensitivity and specificity, which was an average t of 0.047±0.002. Imputed models displayed the following metrics (mean±SD): AUC of 0.86±0.004, sensitivity of 0.80±0.008 and specificity of 0.78±0.008. Metrics and crude ORs of multivariable logistic regression fit to complete case data can be found in online supplemental table S1.

    In this study, we sought to compare the predictability of previously published PRESTO models for paediatric populations in an HIC setting5 and in an LMIC setting6 with a model built with our sample from Tanzania. The coefficients and thresholds published for the afore-mentioned populations were used to assemble these models. Table 4 shows the performance of the imputed model fit to our population and the reconstructed models in predicting our data. Our model built after multiple imputation of missing values performs in close proximity to the model proposed by St. Louis et al5

    View this table:
    Table 4

    Comparison of the Pediatric Resuscitation and Trauma Outcome (PRESTO) models’ performance in predicting in-hospital mortality in paediatric injury patients in Tanzania

    Discussion

    This is the first validation of a model to predict in-hospital mortality for paediatric injury patients in Tanzania. Trauma scores have been developed and used in HICs for triage, trauma registry audit and quality improvement.7 Many of these scores have been used in LMICs, but no single injury severity measure has shown consistent results in all settings.14 Importantly, some scores require data from diagnostic methods such as labs and imaging that are often not available in LMIC settings.15 Thus, many studies have called for adaptations or alternative scoring systems for LMIC populations, with subsequent validation studies.6 14 16 The PRESTO model was developed to predict mortality in children after injury in LMIC settings and includes simple variables available at presentation and on initial resuscitation. In our study, these simple clinical variables needed for the calculation of PRESTO were available in the majority (87%) of patients in our paediatric trauma registry. This was also found in Traynor et al, who applied PRESTO to South African trauma data, and represents a strength of this scoring system for use in LMICs.7

    An additional strength of the PRESTO model above other trauma severity scores, including the Revised Trauma Score17 and the Kampala Trauma Score,18 is that it was developed specifically for children. There are not many trauma scores that have been specifically validated in children. Other studies in HICs and LMICs have found that the PRESTO model is the superior tool to predict mortality for paediatric injury patients in LMICs.5–7 It has been found to outperform the Kampala Trauma Score in children<5 years of age in Rwanda.6 Another prediction score, the Trauma and Injury Severity Score, performs well in older children but not children less than 5 years old.19 20 Thus, PRESTO fills a needed gap in the literature for predicting mortality in injured children in LMIC settings.

    Our study aimed to validate the PRESTO model in another LMIC setting and is the first validation of a model to predict in-hospital mortality for paediatric injury patients in Tanzania. The PRESTO model applied to our population in Tanzania showed good predictive potential for in-hospital mortality. Our findings support those of St-Louis et al, who found that the PRESTO model can predict in-hospital mortality based on bedside variables available on initial patient assessment and resuscitation.6

    In applying the PRESTO model coefficients that were validated in studies using data from the USA and Rwanda to our study population, we see that our Tanzanian imputed model performed similarly. In calibrating the model coefficients for our population, we were able to derive a model that has good predictive potential. Despite our limited sample size, our model displays similar performance to the validated models from the USA and Rwanda that both had large sample sizes.5 6 Thus, this analysis found that the PRESTO model fit to our population in Tanzania showed good predictive potential for in-hospital mortality among children in our paediatric trauma registry. Given this, the PRESTO model can be used in this population for research and quality improvement in order to predict which patients might benefit most from an intervention to improve outcomes. In addition, the PRESTO model could be used as part of a decision support tool at an outside facility to assist medical providers make transfer decisions. If a patient has a high risk of mortality based on the PRESTO Score, the provider may choose to transfer them to a higher level of care.

    Regarding the variables studied, five were found to be statistically associated with in-hospital mortality in the univariate model. When controlling for confounding effects with a multivariable model, only neurologic status (AVPU) and need for SO remained significant. We believe that this change might be due to collinearity found among variables. Although other associations might be in place, need for SO had the strongest association with mortality (OR 13.0) and thus other variables such as HR and hypotension lost their statistical power.

    Limitations

    This study had a relatively limited sample size with only 33 deaths. However, despite this, our model still performs well in predicting mortality and performs similarly to models published previously with larger sample sizes.5 6 Further, the data used in this study is from a registry that is hospital based, and thus we are missing data prior to hospital presentation including those who die prior to presentation at KCMC. As a result, the real burden of paediatric injuries in Northern Tanzania is not known, and expanding the registry to include a community component would likely result in a more accurate picture of the extent of paediatric injuries in this region. Additionally, we did have missing data in completing the PRESTO model in our population (n=67 with missing variables). However, we performed multiple imputation methods to deal with this missing data and feel that our conclusions are sound.

    Conclusion

    In conclusion, in applying the PRESTO model to our population, our model showed good predictive potential for in-hospital mortality. Further research with a larger injury population should be done to see if we can improve the model for our population in Tanzania, such as through calibration. In addition, data from other LMICs should be included and further validation is required in other LMICs to increase the external validity of PRESTO.

    Data availability statement

    The data for this manuscript is covered under a data sharing agreement, and thus we are unable to openly share these data. For data access requests, please contact our non-author KCMC representative Gwamaka William at gwamakawilliam14{at}gmail.com.

    Ethics statements

    Patient consent for publication

    Ethics approval

    This study was approved by the institutional review boards at the Tanzanian National Institute for Medical Research (NIMR/HQ/R.8a/Vol. IX/3475), Kilimanjaro Christian Medical Centre (1252) and the University of Utah (IRB_00134560).

    Acknowledgments

    The authors would like to acknowledge the children of Tanzania in our study and their families.

    References

    1. ↵
      1. GBD 2017 Causes of Death Collaborators
      . Global, regional, and national age-sex-specific mortality for 282 causes of death in 195 countries and territories, 1980-2017: a systematic analysis for the global burden of disease study 2017. Lancet 2018;392:1736–88. doi:10.1016/S0140-6736(18)32203-7
      OpenUrlCrossRefPubMed
    2. ↵
      1. Pérez Méndez MA,
      2. Kigwangalla HA,
      3. Bärnighausen T, et al
      . Injuries among children and adolescents in a rapidly growing urban African Metropolis: a cross-sectional survey of 1,968 households in Dar ES Salaam, Tanzania. PeerJ 2020;8:e10048. doi:10.7717/peerj.10048
    3. ↵
      1. Al-Hajj S,
      2. El Bcheraoui C,
      3. Daoud F, et al
      . Child and adolescent injury burden in the eastern Mediterranean region: findings from the global burden of disease 1990-2017. BMC Public Health 2020;20:433. doi:10.1186/s12889-020-08523-w
    4. ↵
      1. Tupetz A,
      2. Friedman K,
      3. Zhao D, et al
      . Prevention of childhood unintentional injuries in low- and middle-income countries: a systematic review. PLoS ONE 2020;15:e0243464. doi:10.1371/journal.pone.0243464
    5. ↵
      1. St-Louis E,
      2. Bracco D,
      3. Hanley J, et al
      . Development and validation of a new pediatric resuscitation and trauma outcome (PRESTO) model using the U.S. National trauma data bank. J Pediatr Surg 2017:S0022-3468(17)30661-9. doi:10.1016/j.jpedsurg.2017.10.039
    6. ↵
      1. St-Louis E,
      2. Petroze R,
      3. Baird R, et al
      . Calibration and validation of the pediatric resuscitation and trauma outcome model among injured children in rwanda. J Pediatr Surg 2020;55:2510–6. doi:10.1016/j.jpedsurg.2020.01.056
      OpenUrl
    7. ↵
      1. Traynor MD,
      2. St Louis E,
      3. Hernandez MC, et al
      . Comparison of the pediatric resuscitation and trauma outcome (PRESTO) model and pediatric trauma scoring systems in a middle-income country. World J Surg 2020;44:2518–25. doi:10.1007/s00268-020-05512-3
      OpenUrl
    8. ↵
      1. Keating EM,
      2. Sakita F,
      3. Mmbaga BT, et al
      . A cohort of pediatric injury patients from a hospital-based trauma registry in northern Tanzania. Afr J Emerg Med 2022;12:208–15. doi:10.1016/j.afjem.2022.04.008
      OpenUrl
    9. ↵
      1. Harris PA,
      2. Taylor R,
      3. Minor BL, et al
      . The redcap consortium: building an international community of software platform partners. J Biomed Inform 2019;95:103208. doi:10.1016/j.jbi.2019.103208
      OpenUrlCrossRefPubMed
    10. ↵
      1. Harris PA,
      2. Taylor R,
      3. Thielke R, et al
      . Research electronic data capture (redcap) -- a metadata-driven methodology and workflow process for providing translational research informatics support. J Biomed Inform 2009;42:377–81. doi:10.1016/j.jbi.2008.08.010
      OpenUrlCrossRefPubMedWeb of Science
    11. ↵
      1. Nuttall AG,
      2. Paton KM,
      3. Kemp AM
      . To what extent are GCS and AVPU equivalent to each other when assessing the level of consciousness of children with head injury? A cross-sectional study of UK hospital admissions. BMJ Open 2018;8:e023216. doi:10.1136/bmjopen-2018-023216
    12. ↵
      1. Team RC
      . R: A language and environment for statistical computing. 2022.
    13. ↵
      1. van S,
      2. Groothuis-Oudshoorn K
      . Mice multivariate imputation by chained equations in R. J Stat Softw 2011;45. doi:10.18637/jss.v045.i03
    14. ↵
      1. Mehmood A,
      2. Hung YW,
      3. He H, et al
      . Performance of injury severity measures in trauma research: a literature review and validation analysis of studies from low-income and middle-income countries. BMJ Open 2019;9:e023161. doi:10.1136/bmjopen-2018-023161
    15. ↵
      1. St-Louis E,
      2. Séguin J,
      3. Roizblatt D, et al
      . Systematic review and need assessment of pediatric trauma outcome benchmarking tools for low-resource settings. Pediatr Surg Int 2017;33:299–309. doi:10.1007/s00383-016-4024-9
      OpenUrl
    16. ↵
      1. Feldhaus I,
      2. Carvalho M,
      3. Waiz G, et al
      . Thefeasibility, appropriateness, and applicability of trauma scoring systems in low and middle-income countries: a systematic review. Trauma Surg Acute Care Open 2020;5:e000424. doi:10.1136/tsaco-2019-000424
    17. ↵
      1. Champion HR,
      2. Copes WS,
      3. Sacco WJ, et al
      . A new characterization of injury severity. The Journal of Trauma: Injury, Infection, and Critical Care 1990;30:539–46. doi:10.1097/00005373-199005000-00003
      OpenUrl
    18. ↵
      1. Weeks SR,
      2. Juillard CJ,
      3. Monono ME, et al
      . Is the Kampala trauma score an effective predictor of mortality in low-resource settings? A comparison of multiple trauma severity scores. World J Surg 2014;38:1905–11. doi:10.1007/s00268-014-2496-0
      OpenUrlPubMed
    19. ↵
      1. Toida C,
      2. Muguruma T,
      3. Gakumazawa M, et al
      . Validation of age-specific survival prediction in pediatric patients with blunt trauma using trauma and injury severity score methodology: a ten-year nationwide observational study. BMC Emerg Med 2020;20:91.:91. doi:10.1186/s12873-020-00385-0
      OpenUrl
    20. ↵
      1. Schall LC,
      2. Potoka DA,
      3. Ford HR
      . A new method for estimating probability of survival in pediatric patients using revised TRISS methodology based on age-adjusted weights. J Trauma 2002;52:235–41. doi:10.1097/00005373-200202000-00006
      OpenUrlPubMedWeb of Science

    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 EMK contributed to the conceptualisation, data curation, funding acquisition, investigation, methodology, project administration, visualisation and writing—original draft. EMK accepts full responsibility for the work and/or the conduct of the study, had access to the data, and controlled the decision to publish.

      MM contributed to the conceptualisation, methodology, visualisation and writing—original draft. AK contributed to the conceptualisation, methodology, visualisation and writing—original draft. JVS contributed to the conceptualisation, data curation, formal analysis, methodology, software, validation, visualisation and writing—original draft. JVS has directly accessed and verified the underlying data reported in the manuscript. CSA contributed to the conceptualisation, methodology, visualisation and writing—original draft. DBC contributed to the conceptualisation, data curation, formal analysis, methodology, software, validation, visualisation and writing—review & editing. DBC has directly accessed and verified the underlying data reported in the manuscript. CS contributed to the conceptualisation, funding acquisition, investigation, methodology, project administration, supervision and writing—review & editing. BM contributed to the funding acquisition, investigation, project administration, resources, supervision and writing—review & editing. JRNV contributed to the conceptualisation, data curation, formal analysis, methodology, software, supervision, validation, visualisation and writing—review & editing. JRNV has directly accessed and verified the underlying data reported in the manuscript. All authors had full access to the data in the study and accept responsibility to submit for publication.

    • Funding This work was supported by the Fogarty International Center of the National Institutes of Health (D43 TW009337). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The funding source had no involvement in the study design; the collection, analysis and interpretation of data; the writing of the report; or in the decision to submit the article for publication.

    • 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.