Hostname: page-component-586b7cd67f-tf8b9 Total loading time: 0 Render date: 2024-11-27T22:41:11.911Z Has data issue: false hasContentIssue false

Developing a novel risk prediction model for severe malarial anemia

Published online by Cambridge University Press:  11 September 2017

E. B. Brickley
Affiliation:
Laboratory of Malaria Immunology and Vaccinology, Division of Intramural Research, National Institute of Allergy and Infectious Diseases, National Institutes of Health, Rockville, Maryland, USA Department of Public Health and Primary Care, University of Cambridge, Cambridge, UK Department of Epidemiology, Geisel School of Medicine at Dartmouth College, Lebanon, New Hampshire, USA
E. Kabyemela
Affiliation:
Muheza Designated District Hospital, Muheza, Tanzania
J. D. Kurtis
Affiliation:
Department of Pathology and Laboratory Medicine, Rhode Island Hospital, Brown University Medical School, Providence, Rhode Island, USA
M. Fried
Affiliation:
Laboratory of Malaria Immunology and Vaccinology, Division of Intramural Research, National Institute of Allergy and Infectious Diseases, National Institutes of Health, Rockville, Maryland, USA
A. M. Wood
Affiliation:
Department of Public Health and Primary Care, University of Cambridge, Cambridge, UK
P. E. Duffy*
Affiliation:
Laboratory of Malaria Immunology and Vaccinology, Division of Intramural Research, National Institute of Allergy and Infectious Diseases, National Institutes of Health, Rockville, Maryland, USA
*
*Address for correspondence: P. E. Duffy, Laboratory of Malaria Immunology and Vaccinology, NIAID, NIH, Twinbrook I, Room 1111, 5640 Fishers Lane, Rockville, MD 20852, USA (Email: [email protected])
Rights & Permissions [Opens in a new window]

Abstract

As a pilot study to investigate whether personalized medicine approaches could have value for the reduction of malaria-related mortality in young children, we evaluated questionnaire and biomarker data collected from the Mother Offspring Malaria Study Project birth cohort (Muheza, Tanzania, 2002–2006) at the time of delivery as potential prognostic markers for pediatric severe malarial anemia. Severe malarial anemia, defined here as a Plasmodium falciparum infection accompanied by hemoglobin levels below 50 g/L, is a key manifestation of life-threatening malaria in high transmission regions. For this study sample, a prediction model incorporating cord blood levels of interleukin-1β provided the strongest discrimination of severe malarial anemia risk with a C-index of 0.77 (95% CI 0.70–0.84), whereas a pragmatic model based on sex, gravidity, transmission season at delivery, and bed net possession yielded a more modest C-index of 0.63 (95% CI 0.54–0.71). Although additional studies, ideally incorporating larger sample sizes and higher event per predictor ratios, are needed to externally validate these prediction models, the findings provide proof of concept that risk score-based screening programs could be developed to avert severe malaria cases in early childhood.

Type
Original Research Article
Creative Commons
Creative Common License - CCCreative Common License - BY
This is a work of the U.S. Government and is not subject to copyright protection in the United States. This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted re-use, distribution, and reproduction in any medium, provided the original work is properly cited
Copyright
Copyright © The Author(s) 2017

Introduction

Although improved uptake of vector control measures and first-line antimalarial medication has reduced the number of deaths due to malaria in children younger than 5 years of age by nearly 60% since the year 2000, malaria still remains the fourth leading cause of child death in sub-Saharan Africa [Reference Bhatt1, 2]. In the Global Technical Strategy for Malaria 2016–2030 endorsed in May 2015, the World Health Assembly set the goal of reducing the 2015 malaria mortality rates by a further 90% by 2030 [3]. Mathematical models of Plasmodium falciparum transmission suggest that scaling-up of existing population-wide interventions, including through expanded access to long-lasting insecticidal nets (LLINs), seasonal malaria chemoprevention in children aged 6 months to 5 years, and artesunate therapies, could reduce malaria mortality rates by up to an estimated 74% (95% CI 67–84%) [Reference Griffin4]. Further, the incorporation of ‘near-term innovations,’ such as LLINs with four-year half-lives, broader use of seasonal malaria chemoprevention in children aged 6–10 years, and the use of alternative chemopreventive medications in East African regions with high sulfadoxine-pyrimethamine parasitic resistance, could lead to mortality reductions of up to 81% (95% CI 76–87%) [Reference Griffin4]. By using population-level strategies that seek to control the causes of malarial death, the proposed scenarios could potentially save millions of lives. Nevertheless, implementation of these agendas will be highly costly and, barring development of the near-term innovations, will fall short of achieving the 90% mortality reduction benchmark. Therefore, we asked: Would it be feasible to complement the ‘population approach’ with a ‘high-risk approach’ that aims to provide additional targeted intervention to the children who are most susceptible to developing life-threatening malaria? [Reference Rose5]

Using data collected at the time of birth as part of the Mother Offspring Malaria Study (MOMS) Project birth cohort in Muheza, United Republic of Tanzania, between 2002 and 2006, we sought to develop a prognostic model for severe malarial anemia as a pilot study for investigating whether personalized medicine approaches could have value for the reduction of malaria-related mortality. In regions with high malaria transmission, severe malarial anemia, here defined as a P. falciparum infection with hemoglobin concentrations below 50 g/L, is the most common manifestation of life-threatening malaria in young children [Reference O'Meara6]. Moreover, previous studies on this cohort provided evidence that cord levels of the pro-inflammatory cytokines tumor necrosis factor (TNF) and interleukin-1β (IL-1β) may correlate with children's cytokine levels throughout infancy and may be inversely related to future parasite burdens and severe malarial anemia risk [Reference Kabyemela7, Reference Brickley8]. Building on this research, we aimed to: (i) assess the discriminative ability of subject characteristics and biological markers measured at delivery for the prediction of a child's probability of developing severe malarial anemia during the first 3 years of life and (ii) estimate the potential clinical benefit of using a risk prediction model to guide further intervention.

Methods

Study cohort

The study setting, location, dates, and methods for data collection and follow-up in the MOMS Project have been described in detail previously [Reference Goncalves9]. To be eligible for the study, children had to be: (i) born to human immunodeficiency virus negative mothers, (ii) sickle cell disease free, (iii) singleton births, and (iv) followed for a minimum of 28 days. The primary outcome was time to first severe malarial anemia. Clinical teams monitored children for severe malarial anemia at all sick visits and during routine visits occurring every 2 weeks in infancy and every 4 weeks thereafter until age 4 or study termination. Parasitemia was assessed by Giemsa-stained thick blood smear, and hemoglobin levels were measured with an impedance-based analyzer. The candidate predictors included: sex, genotypes for beta-globin (i.e., AA and AS) and alpha-thalassemia (i.e., α/α, α/α − 3.7, and α − 3.7/α − 3.7), birth weight, maternal age, gravidity (i.e., here indicated by number of previous pregnancies), transmission season at delivery (i.e., early high season (May–July), late high season (August–October), early low season (November–January), and late low season (February–April), placental malaria status at delivery (i.e., the detection of P. falciparum parasitemia in placental blood collected through mechanical pressing of full-thickness tissue), household bed net possession at delivery (i.e., treated, untreated, and no/unknown), intermittent preventive treatment in pregnancy (IPTp) (i.e., 0 versus 1+ doses), and a panel of cord blood cytokines and receptors listed in Table 1 and visualized in Supplementary Fig. S1. Sickle cell trait was determined by cellulose acetate paper electrophoresis (Helena Laboratories, Beaumont, TX, USA). Alpha-thalassemia was determined by PCR [Reference Chong10]. Birth weight was evaluated within 24 h after delivery. Data on maternal age, gravidity, household bed net possession, and IPTp dosage were self-reported. Soluble cytokines and receptors in cord blood plasma were measured using commercially available multiplex, bead-based platforms (BioPlex®, BioRad, Irvine, CA, USA) and custom-made assay kits as previously described [Reference Coutinho11]. Samples that did not produce detectable concentrations of a given marker were assigned a value of half the limit of detection of that marker. As the numbers of measurements detectable by immunoassays were low for interleukin-4 (9.7%, n = 76) and interferon-γ (20.6%, n = 161), these cytokines were evaluated as binary variables (i.e., detectable levels versus levels below the limits of detection); other cord cytokines and receptors were loge-transformed to obtain approximately normal distributions (Supplementary Fig. S1).

Table 1. Distribution of baseline characteristics in the MOMS Project (2002–2006) birth cohort in Muheza, Tanzania

Study oversight

This study was approved by the US National Institutes of Health International Clinical Studies Review Committee of the Division of Microbiology and Infectious Diseases. Ethical clearance was provided by the Institutional Review Boards of the Seattle Biomedical Research Institute and the National Institute for Medical Research in Tanzania. Participating mothers provided written informed consent for themselves and their newborn child. Prompt care was provided to sick children in accordance with Tanzanian Ministry of Health protocols. All subsequent laboratory measurements were performed on de-identified samples.

Statistical analysis

Associations between the baseline risk markers and severe malarial anemia were quantified using Cox proportional hazard models. Candidate markers were investigated first individually using all of the available data for a given marker and then in combined analyses restricted to those individuals with complete information. A basic model using the pragmatic markers of sex, gravidity, transmission season at delivery, and bed net possession was also explored. Risk discrimination was assessed using Harrell's C-index, a generalization of the area under the receiver operating curve suitable for time-to-event analyses [Reference Harrell12]. The C-index quantifies the ability of a model to correctly predict the order of disease events and ranges in value from 0.5 (i.e., chance) to 1 (i.e., perfect prediction of order failure). To investigate the magnitude of optimism (i.e., overfitting of the model due to the small number of cases and high number of predictors) [Reference Smith13], bootstrapping was performed using the technique originally described by Harrell et al. [Reference Harrell, Lee and Mark14]. For this method, the data set was resampled to produce 200 replicated data sets. Then, the prognostic models were fitted to each of the 200 bootstrapped data sets and the original data set. The C-indices were calculated for all data sets, and the difference between each resampled data set and the original data set was recorded. Estimates of optimism were calculated as the average of the 200 differences. To correct the models, the optimism estimates were then subtracted from each of the original C-indices. To elucidate the extent to which risk prediction might have benefit for public health campaigns against malaria, a modeling exercise was performed to explore a hypothetical screening program with targeted versus randomized allocation of antimalarial intervention. Specifically, it was assumed that the population had 1000 children, an incidence of severe malarial anemia equivalent to that observed in the subset of the MOMS Project cohort with available cord blood cytokine measurements (n = 781) who were followed between 2002 and 2006, and resources available to administer an intervention with 75% protective efficacy [Reference Wilson and Taskforce15, Reference Meremikwu16] to 200 of the children. All statistical analyses were performed using Stata – version 12 (StataCorp LP, College Station, TX, USA).

Results

During a median follow-up of 2.0 (interquartile range, 1.1–2.9) years, 37 incident severe malarial anemia events were recorded in the 880 children participating in the MOMS Project cohort. Out of the baseline demographic and clinical markers, the transmission season at delivery (evaluated quarterly) provided the best discrimination of severe malarial anemia risk, and its associated C-index was 0.63 (95% CI 0.54–0.71) (Fig. 1). Household bed net possession at enrollment was also significantly predictive in univariate models with a C-index of 0.61 (95% CI 0.53–0.68) (Fig. 1). Combining these two predictors with information on the child's sex and the mother's gravidity that are readily obtainable at the time of birth yielded an additive C-index of 0.69 (95% CI 0.61–0.77). However, the estimated optimism for this prognostic model was substantial, and after performing a bootstrap correction, the C-index remained significant but attenuated to 0.63 (95% CI 0.55–0.71) (Fig. 2).

Fig. 1. Discrimination of severe malarial anemia risk in univariate prognostic models in the MOMS Project (2002– 2006) birth cohort in Muheza, Tanzania, ranked in order of increasing C-index estimates. IPTP, intermittent preventive treatment in pregnancy; IFN, interferon; IL, interleukin; TNF-RI, tumor necrosis factor-receptor I; TNF, tumor necrosis factor; TNF-RII, tumor necrosis factor-receptor II.

Fig. 2. Optimism-corrected C-indices for severe malarial anemia risk in the MOMS Project (2002–2006) birth cohort in Muheza, Tanzania. Abbreviations: IL, interleukin.

Out of the cord blood markers, interleukin-1β yielded the strongest discrimination of risk with a C-index of 0.77 (95% CI 0.70–0.84) and had an average optimism of only 0.0044 (Figs 1 and 2). In a hypothetical cord blood screening program that measures interleukin-1β in 1000 children, the targeted allocation of an intensive antimalarial intervention with 75% efficacy to the top 20% of the population at risk of severe malarial anemia (i.e., the high risk group) would avert 13 additional severe malarial anemia events compared with the distribution of the intervention to a randomly selected sample of 20% of the population (Fig. 3). Stratifying the population's risk with this 20% cut-off yielded a ROC area of 0.69 (95% CI 0.61–0.78) (Table 2). Given the total population prevalence of severe malarial anemia of 4.7% (95% CI 3.4–6.5%), the positive predictive value (i.e., the probability that children who received the intensive intervention developed severe malarial anemia) was enriched to 13.5% (95% CI 8.5–19.8%) in the high risk group, and the negative predictive value (i.e., the probability that children who did not receive the intensive intervention did not develop severe malarial anemia) remained high at 97.4% (95% CI 95.9–98.5%) (Table 2). In this modeled scenario, the cord blood of 77 (i.e., 1000÷13) children would have to be screened for interleukin-1β to avert one additional severe malarial anemia case.

Fig. 3. Schematic diagram for hypothetical interleukin-1β screening program with targeted allocation of antimalarial intervention.

Table 2. Cord blood interleukin-1β risk scores for severe malarial anemia in the MOMS Project (2002–2006) birth cohort in Muheza, Tanzania (n = 781)

The total population prevalence of severe malarial anemia during the first 3 years of life was 4.7% (95% CI 3.4–6.5%). For estimation of sensitivity, specificity, ROC area, and positive and negative predictive value, the lower limit of each risk score quintile was used to define a ‘positive’ result of the screening test (e.g., Q4, 4.4% predicted probability or higher = positive). Bolded text indicates the performance of the high risk group cut-off (i.e., top 20%) used in the public health modeling exercise illustrated in Fig. 3.

Discussion

These data provide proof of concept that a prognostic model based on measurements captured at the time of birth can be developed to facilitate the selection of high-risk children for additional targeted intervention against malaria. Despite the small sample size and number of cases in this pilot study, the prognostic associations, particularly with cord blood cytokines, were strong and had C-indices in the range of those observed for models of HIV acquisition [Reference Kagaayi17] and mortality from severe sepsis [Reference Ricciuto18]. Even in the absence of blood-based measurements, a simple model based on factors readily obtainable at the time of delivery (i.e., sex, gravidity, transmission season at delivery, and bed net possession) was able to correctly predict the severe malarial anemia order failure for 63 out of 100 pairs of children – 13 more pairs than chance alone. Further research is needed to externally validate the presented prognostic models of severe malarial anemia risk and to explore the feasibility of incorporating targeted interventions for high-risk individuals into ongoing antimalarial campaigns in developing countries.

The current study had strengths and limitations. In line with the recommendations of the TRIPOD statement [Reference Collins19], candidate predictors were selected to be evaluable both objectively (and thereby testable in validation cohorts) and at the time of delivery (i.e., before children would be subject to interventions that might alter their exposure to Plasmodium parasites). Minimizing the potential for detection bias, the outcome of severe malarial anemia was independently diagnosed by clinical teams prior to the selection of potential risk predictors. Another strength of this analysis was that associations between the predictors and outcome were internally validated using a bootstrap resampling approach to investigate and correct for overfitting that could arise from the low event to predictor ratio [Reference Smith13]. Moving forward, it will be important to externally validate the prognostic models for severe malarial anemia in other birth cohorts. In particular, the generalizability of the described prediction models may be limited, and it will be important to validate the models using data from regions with different ecological patterns of malaria transmission (e.g., in regions with low or highly seasonal transmission) where the age of severe malarial anemia onset may be delayed [Reference Roca-Feltrer20]. In addition, future studies should aim to recruit larger sample sizes with greater numbers of severe malaria cases in order to mitigate optimistic model performance estimates. Including more participants will also enable investigations of the potential heterogeneity in the model's discrimination that may exist within clinically relevant sub-groups (e.g., in children exposed to placental malaria in utero).

In regions that are approaching the elimination of malaria, identifying high risk groups based on easily ascertainable and low cost indicators will be particularly relevant, and efforts should be made to enhance the feasibility of malaria screening programs. Specifically, future studies that aim to formally develop prognostic models of life-threatening malaria in childhood should evaluate additional pragmatic predictors, such as parental occupation (e.g., working in the logging industry [Reference de Castro21]) and housing conditions (e.g., the presence of window and door screens) [Reference Kirby22, Reference Tusting23]. Investigations into the cost-effectiveness of a potential screening program should also consider alternative assays for evaluating children's blood-based biomarkers (e.g., use of dried blood spots [Reference Skogstrand24]) as well as factors that could influence the context-specific protective efficacy of interventional programs (e.g., relative abundance of drug-resistant parasites [Reference Griffin25]).

In conclusion, this study provides valuable new evidence that the development of a prognostic model for severe malarial anemia during infancy and early childhood is scientifically plausible. By providing clinicians and families with an objectively estimated probability of a newborn's risk of developing life-threatening complications of infection, personalized medicine approaches have the potential to complement on-going population-wide approaches for malaria prevention.

Supplementary Material

To view supplementary material for this article, please visit https://doi.org/10.1017/gheg.2017.8

Acknowledgements

We thank the families who have participated in this study.

This work was supported by the Intramural Research Program of the National Institute of Allergy and Infectious Diseases, National Institutes of Health [http://www.niaid.nih.gov/], the Bill & Melinda Gates Foundation [grant number 29202, http://www.gatesfoundation.org/], the Grand Challenges in Global Health Initiative [grant number 1364, http://grandchallenges.org/], the National Institutes of Health Fogarty International Center [grant number D43 TW005509, http://www.fic.nih.gov/], and the National Institutes of Health [grant number R01A152059 to PED]. During the duration of study, EBB was supported by the National Institutes of Health Oxford Cambridge Scholars Program [http://oxcam.gpp.nih.gov/] and the National Cancer Institute, National Institutes of Health [grant number R25CA134286, https://www.cancer.gov/].

Conflict of Interest

The authors declare no conflicts of interest.

Ethical Standards

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008.

References

1. Bhatt, S, et al. The effect of malaria control on Plasmodium falciparum in Africa between 2000 and 2015. Nature 2015; 526(7572): 207211.CrossRefGoogle ScholarPubMed
2. World Health Organization. World Malaria Report 2015. World Health Organization, Geneva, 2015.Google Scholar
3. WHO. Global Technical Strategy for Malaria 2016–2030. World Health Organization, United Kingdom, 2015.Google Scholar
4. Griffin, JT, et al. Potential for reduction of burden and local elimination of malaria by reducing Plasmodium falciparum malaria transmission: a mathematical modelling study. Lancet Infectious Diseases 2016; 16(4): 465472.CrossRefGoogle ScholarPubMed
5. Rose, G. Sick individuals and sick populations. International Journal of Epidemiology 1985; 14(1): 3238.Google Scholar
6. O'Meara, WP, et al. Effect of a fall in malaria transmission on morbidity and mortality in Kilifi, Kenya. Lancet 2008; 372(9649): 15551562.Google Scholar
7. Kabyemela, E, et al. Cytokine profiles at birth predict malaria severity during infancy. PLoS ONE 2013; 8(10): e77214.CrossRefGoogle ScholarPubMed
8. Brickley, EB, et al. Fetal origins of malarial disease: cord blood cytokines as risk markers for pediatric severe malarial anemia. Journal of Infectious Diseases 2015; 211(3): 436444.Google Scholar
9. Goncalves, BP, et al. Parasite burden and severity of malaria in Tanzanian children. New England Journal of Medicine 2014; 370(19): 17991808.CrossRefGoogle ScholarPubMed
10. Chong, SS, et al. Single-tube multiplex-PCR screen for common deletional determinants of alpha-thalassemia. Blood 2000; 95(1): 360362.Google Scholar
11. Coutinho, HM, et al. Nutritional status and serum cytokine profiles in children, adolescents, and young adults with Schistosoma japonicum-associated hepatic fibrosis, in Leyte, Philippines. Journal of Infectious Diseases 2005; 192(3): 528536.Google Scholar
12. Harrell, FE Jr, et al. Evaluating the yield of medical tests. JAMA 1982; 247(18): 25432546.Google Scholar
13. Smith, GC, et al. Correcting for optimistic prediction in small data sets. American Journal of Epidemiology 2014; 180(3): 318324.CrossRefGoogle ScholarPubMed
14. Harrell, FE Jr, Lee, KL, Mark, DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Statistics in Medicine 1996; 15(4): 361387.Google Scholar
15. Wilson, AL, Taskforce, IP. A systematic review and meta-analysis of the efficacy and safety of intermittent preventive treatment of malaria in children (IPTc). PLoS ONE 2011; 6(2): e16976.Google Scholar
16. Meremikwu, MM, et al. Intermittent preventive treatment for malaria in children living in areas with seasonal transmission. Cochrane Database of Systematic Reviews 2012; 2: CD003756.Google Scholar
17. Kagaayi, J, et al. Indices to measure risk of HIV acquisition in Rakai, Uganda. PLoS ONE 2014; 9(4): e92015.Google Scholar
18. Ricciuto, DR, et al. Angiopoietin-1 and angiopoietin-2 as clinically informative prognostic biomarkers of morbidity and mortality in severe sepsis. Critical Care Medicine 2011; 39(4): 702710.Google Scholar
19. Collins, GS, et al. Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD): the TRIPOD statement. Annals of Internal Medicine 2015; 162(1): 5563.Google Scholar
20. Roca-Feltrer, A, et al. The age patterns of severe malaria syndromes in sub-Saharan Africa across a range of transmission intensities and seasonality settings. Malaria Journal 2010; 9: 282.Google Scholar
21. de Castro, MC, et al. Malaria risk on the Amazon frontier. Proceedings of National Academy of Sciences of the United States of America 2006; 103(7): 24522457.CrossRefGoogle ScholarPubMed
22. Kirby, MJ, et al. Effect of two different house screening interventions on exposure to malaria vectors and on anaemia in children in The Gambia: a randomised controlled trial. Lancet 2009; 374(9694): 9981009.CrossRefGoogle ScholarPubMed
23. Tusting, LS, et al. Housing Improvements and Malaria Risk in Sub-Saharan Africa: a Multi-Country Analysis of Survey Data. PLoS Medicine 2017; 14(2): e1002234.CrossRefGoogle ScholarPubMed
24. Skogstrand, K, et al. Effects of blood sample handling procedures on measurable inflammatory markers in plasma, serum and dried blood spot samples. Journal of Immunological Methods 2008; 336(1): 7884.Google Scholar
25. Griffin, JT, et al. Protective efficacy of intermittent preventive treatment of malaria in infants (IPTi) using sulfadoxine-pyrimethamine and parasite resistance. PLoS ONE 2010; 5(9): e12618.Google Scholar
Figure 0

Table 1. Distribution of baseline characteristics in the MOMS Project (2002–2006) birth cohort in Muheza, Tanzania

Figure 1

Fig. 1. Discrimination of severe malarial anemia risk in univariate prognostic models in the MOMS Project (2002– 2006) birth cohort in Muheza, Tanzania, ranked in order of increasing C-index estimates. IPTP, intermittent preventive treatment in pregnancy; IFN, interferon; IL, interleukin; TNF-RI, tumor necrosis factor-receptor I; TNF, tumor necrosis factor; TNF-RII, tumor necrosis factor-receptor II.

Figure 2

Fig. 2. Optimism-corrected C-indices for severe malarial anemia risk in the MOMS Project (2002–2006) birth cohort in Muheza, Tanzania. Abbreviations: IL, interleukin.

Figure 3

Fig. 3. Schematic diagram for hypothetical interleukin-1β screening program with targeted allocation of antimalarial intervention.

Figure 4

Table 2. Cord blood interleukin-1β risk scores for severe malarial anemia in the MOMS Project (2002–2006) birth cohort in Muheza, Tanzania (n = 781)

Supplementary material: File

Brickley et al supplementary material 1

Supplementary Figure

Download Brickley et al supplementary material 1(File)
File 41.4 KB