Wasting X-Factor

Risk Overview

The x-factor is a risk exposure that tries to capture the differential risk experienced by some children who may experience more relapses of wasting. We believe this is an important component of wasting epidemiology to capture [see Brain Trust notes with Chris Murray for discussion of adding this component to the model]. There are many risk factors that have been described in the literature that pre-dispose children to wasting including maternal education, household food insecurity, family size, water and sanitation.

Differential relapse rate literature overview

[Stobaugh-et-al-2019] conducted a systematic literature review on relapse of SAM following treatment and found that “the proportion of children who relapsed after SAM treatment varied greatly from 0% to 37% across varying lengths of time following discharge” (p. 1). Generally, relapse was defined as presenting with SAM at least once following SAM treatment, but some studies defined relapse as inclusive of MAM as well. Additionally, no standard length of follow-up was used and the period ranged from 1 week to 18 months. [Stobaugh-et-al-2019] reported that relapse tended to occur more frequently in the first 6 months following discharge.

[Stobaugh-et-al-2019] discussed an unpublished longitudinal study in Ethiopia (Tsineal et al. 2015/Jimma University) that was reportedly the first to prospectively examine the risk of acute malnutrition among post-SAM children relative to community controls without a recent history of acute malnutirtion. Notably, this study appears to have since been published by [Girma-et-al-2022].

[Girma-et-al-2022] conducted a prospective matched cohort study of acute malnutrition relapse in Jimma Zone, Ethiopia between September 2013 and September 2015. Inclusion criteria for the post-SAM children were aged 6-59 months at the time of admission into the CMAM program and successful discharge from the program according to the current national guideline (MUAC>11cm, weight gain of 20%, absence of odedema, and clinically stable for two consecutive weeks). Controls were eligible if they were apparently healthy with no history of an acute malnutrition episode and were matched 1:1 to a post-SAM child by age and sex by asking the case’s caretaker to indicate neighboring households with children of the same age and sex. Post-SAM and control children were followed concurrently each month for 12 months. The relevant results of this study are summarized below:

Girma et al. 2022 results

Outcome	Group	No. episodes	Person-time	Incidence rate	Incidence rate ratio
Acute malnutrition	Post-SAM	48	1297	3.7	5.5 (4.7, 15.9)
Acute malnutrition	Control	14	2105	0.7	1
Moderate acute malnutrition	Post-SAM	34	1345	2.5	4.1 (2.1, 8.4)
Moderate acute malnutrition	Control	13	2105	0.6	1
Severe acute malnutrition	Post-SAM	26	2044	1.27	14.1 (3.5, 122.5)
Severe acute malnutrition	Control	2	2216	0.09	1

Since the time of this review, two additional studies were identified that compared wasting episode relapse rates betwen children who were recently treated for acute malnutrition and children who were not.

[Abitew-et-al-2020] conducted a cross-sectional study in South Gondar Zone, Amhara Region, Ethiopia of children discharged from community management of acute malnutrition (CMAM) programs and aged-matched children who were never treated for acute malnutrition. The study found that 34.2% of children discharged from CMAM programs were wasted compared to 26.7% of the control group. More specifically, 22.3% and 12.0% of the children discharged from CMAM were classified as having MAM and SAM, respectively, compared to 17.2% and 9.5% of the control population. Notably, the difference in wasting burden between these study populations found by [Abitew-et-al-2020] is of lower magnitude than other data sources that investigated the same measure. This may be because of the cross-sectional rather than longitudinal study design which measures prevalence rather than cumulative incidence. Further, the inclusion criteria for control subjects was based on whether they had previously received CMAM treatment without consideration of CMAM treatment need. Additionally, the prevalence of SAM in this population is quite high, indicating that there is a high background risk of acute malnutrition in the population that may not be generalizable to other populations with lower SAM prevalence.

[Adegok-et-al-2020] performed a prospective matched cohort study of SAM recurrance in Northern Nigeria. The study recruited children discharged from OTP treatment and matched community controls and followed them for 6 months. At the end of follow-up 24% of the OTP discharged children relapsed to SAM and 0.6% of community controls developed SAM.

Additionally, a study protocol for a prospective matched cohort study of acute malnutrition relapse (similar to [Girma-et-al-2022]) conducted in Mali, Somalia, and South Sudan has been published by [King-et-al-2022], with data collection expected to conclude in January of 2023.

Note

Given the limitations associated with the cross-sectional nature of the [Abitew-et-al-2020] study and the study location of Nigeria rather than Ethiopia for the [Adegok-et-al-2020] study, we decided to use the study conducted by [Girma-et-al-2022] as the main data source to inform differential relapse rates by x-factor exposure status in the acute malnutrition simulation.

Additional literature on relapse rates

[Lambebo-et-al-2021] conducted a retrospective records review of children under five years of age who were admitted and discharged for SAM in 20 selected health posts in Hadiya zone, SNNPR, Ethiopia from 2014/2015 to 2019/2020. Notably, treatment defaulters and non-responders were not excluded from the cohort. Additionally, the cohort was open and children were not followed for relapse after the age of 60 months. They reported a relapse/readmission rate of 9.6% (95% CI: 7.7, 11.7) within the five year study period.

[Abitew-et-al-2020b] conducted a study of children 6-59 months of age who were discharged from community-based management of acute malnutrition (CMAM) for MAM and uncomplicated SAM in the South Gonda Zone of Amhara region, Ethiopia. The study reported that after an average of 5.2 months of recovery, 445 of 1,273 children (35%) had relapsed. [Abitew-et-al-2020b] also found higher odds of relapse among children who were male, received pre-lacteal feeding, lived in food insecure households, and were not given vitamin A supplements.

[Chang-et-al-2013] found that 63% of children successfully treated for MAM in Malawi remained well nourished during 12 months of follow-up (7% were lost to follow-up and 10% progressed to SAM). [Stobaugh-et-al-2018] found that 58% of children sucessfully treated for MAM in Malawi reamined well nourished during 12 months of follow-up. [Trehan-et-al-2015] found that 71% and 63% of children treated for MAM (defined as 12 weeks of treatment and until WHZ>-2, respectively) in Malawi did not relapse during 12 months of follow-up.

Vivarium Modeling Strategy

The literature summarized above relates to differential risk of acute malnutrition incidence as it relates to a recent episode of acute malnutrition. We have previously used this estimation of risk effect as a proxy measure for the risk effect of an environmental-type risk factor with an exposure status that does not change over time (like socioeconomic status or food security, for example rather than “episode of acute malnutrition in the past year”) – see the existing corresponding X-factor risk exposure page. Note that an alternative implementation of this risk factor could be paired with a dynamic exposure model that corresponds to an individual simulant’s time since last acute malnutrition episode.

Note that these two alternative approaches are quite different in their implications. A risk factor based on a recent acute malnutrition episode would imply that averting a given case of acute malnutrition could avert furture cases of acute malnutrition whereas the “environmental-type” risk factor approach would not. Note that in reality, both such forces are likely at play and would require careful epidemiologic analysis to disentangle these differential types of causal effects.

In any case, calibrating the baseline wasting exposure model and the x-factor risk factor implementation poses a significant challenge. We have not previously attempted to calibrate a model that uses a dynamic risk exposure implementation of the x-factor risk based on recent acute malnutrition episodes.

We have attempted to calibrate the wasting and x-factor models using a static risk exposure implementation of the x-factor model for the CIFF malnutrition project. We had assumed risk effect values as informed from the literature, but a primary remaining question was how to inform the population-level risk exposure of the x-factor model. For the prior implementation, we ran the simulation under several different population-level exposure values for the x-factor. We wrote a custom observer using the interactive context to observe the proportion of simulants who recovered from SAM/MAM who had a subsequent episode of SAM/MAM in the following year (see this notebook for details on the code). We then selected the population-level exposure that yielded results that were most similar to the relapse rate statistics summarized in the literature above. However, this process was additionally complicated/limited by the need for a burn-in period to reach a steady state across wasting categories (as x-factor exposure will vary by wasting state), lack of age specificity, vague/uncertain validation targets, and process inefficiencies. Note that we had initially considered setting population-level x-factor exposure equal to the prevalence of SAM, but this did not validate well to external validation targets, which was suspected to be a result of the prevalence of SAM being unequal to the proportion of children who experienced an episode of SAM in the past year.

Notably, the x-factor implementation had very little impact on population-level results for the CIFF project. Despite the implementation of the x-factor risk factor being imperfect in concept and practice, further improvement to the risk factor was not pursed given this minimal impact within this model. Inclusion of the x-factor risk is expected to be more impactful if any interventions are targeted to the x-factor (either directly or through some correlated factor), such as targeting SQ-LNS to those who are discharged from acute malnutrition treatment. However, if we plan to include the x-factor risk in a model in the future, additional thought should be put into how to operationalize the x-factor exposure model as well as best practices for calibrating the x-factor and wasting risk exposure models before proceeding.

Finally – an entirely different conceptualization of the x-factor risk effects could also be considered. For instance, rather than informing x-factor effects through literature related to wasting relapse, we could inform such effects through literature that examines drivers of wasting such as socioeconomic status and/or food security that are more aligned with the static risk exposure implementation and that have directly measureable exposure values. This would resolve a challenge of calibrating the population-level exposure distribution and would leave only the remaining challenge of finding the wasting state-specific exposure distribution of the “x-factor” (which will differ from the population level distribution). However, research goals of the project for which the x-factor risk factor implementation will be applied should guide whether or not this is an appropriate direction.

Wasting Incidence Rates

X-factor risk effects

Affected parameter	Exposure category	Risk effect type	Risk effect	Note
i1	Exposed (cat1)	Relative risk (RR)	3.44 (TODO: implement monte carlo uncertainty estimation about this parameter)	Derived from [Girma-et-al-2022] as \(RR_\text{SAM} / RR_\text{MAM}\). This value was updated from its previous value of \(\sqrt{10}\) following the publication of [Girma-et-al-2022]
i1	Unexposed (cat2)	Relative risk (RR)	\(1\)
i2	Exposed (cat1)	Relative risk (RR)	4.1 (95% CI: 2.1, 8.4, lognormal distribution of uncertainty)	[Girma-et-al-2022]. This value was updated from its previous value of \(\sqrt{10}\) following the publication of [Girma-et-al-2022]
i2	Unexposed (cat2)	Relative risk (RR)	\(1\)
i3	Exposed (cat1)	Relative risk (RR)	\(1\)
i3	Unexposed (cat2)	Relative risk (RR)	\(1\)

For each incidence rate \(i(n)\) in the dynamic wasting exposure model (i1, i2, and i3), the simulant-specific rate should be determined as follows:

\[i(n)_i = i(n) \times (1 - PAF_\text{i(n)}) \times RR_\text{i(n),i}\]

\[PAF_\text{i(n)} = \frac{RR_\text{i(n),cat1} * p_\text{cat1|source wasting state} + RR_\text{i(n),cat2} * (1 - p_\text{cat1|source wasting state}) - 1}{RR_\text{i(n),cat1} * p_\text{cat1|source wasting state} + RR_\text{i(n),cat2} * (1 - p_\text{cat1|source wasting state})}\]

Where:

• \(p_\text{cat1|source wasting state}\) is the x-factor risk exposure among the source state for the relevant transition. Values shown in the table below:

Warning

The values in the table below were specific to a given calibration of the x-factor risk implementation and will need to be re-estimated via a calibration process if the x-factor exposure will be included in any additional models.

X-factor risk exposure by wasting state at initialization

Transition	Source wasting state	\(p_\text{cat1}\) value	Note
i1	moderate acute malnutrition	0.78
i2	mild child wasting	0.54
i3	susceptible to child wasting	0.45	Since the relative risk for this transition is equal to one, the impact of this value should be inconsequential

Note

This custom PAF calculation strategy was chosen due to the underestimation of MAM and overestimation of SAM child wasting exposure states in model version 4.5.5 of the acute malnutrition simulation. This appeared to be a result of an overestimation of the transition rate between MAM to SAM (i1) as a result of the higher x-factor exposure present in the MAM wasting state than the general population, as shown in this notebook.

The values in the table below are the x-factor exposures specific to the source wasting state for each wasting transition, as calculated among ages 6 months to 5 years after the first “burn-in” year of the simulation run.

Note

Additional factors will affect wasting incidence rates i1, i2, and i3 in the acute malnutrition simulation, including diarrheal diseases and SQ-LNS.

Validation and Verification Criteria

• The ratio between wasting incidence rates among those exposed and unexposed to the x-factor should match the given x-factor effect size

• There should be no difference in wasting state remission rates by x-factor exposure status

• Wasting exposure should be greater among those exposed to the x-factor than those unexposed

• Wasting exposure should continue to validate to GBD

Assumptions and Limitations

• We assume wasting recovery rates are not affected by the x-factor

• We do not model a direct causal effect of an episode of wasting on future episodes of wasting

• We are generalizing the effect of the post-SAM recovery state on acute malnutrition incidence to our “x-factor” risk factor that is not necessarily specific to the post-SAM recovery state, but rather is modeled as a constant “vulnerability” risk factor among a subset of modeled children.

References

Abitew-et-al-2020(1,2,3)

Abitew DB, Worku A, Mulugeta A, Bazzano AN. Rural children remain more at risk of acute malnutrition following exit from community based management of acute malnutrition program in South Gondar Zone, Amhara Region, Ethiopia: a comparative cross-sectional study. PeerJ. 2020 Feb 7;8:e8419. doi: 10.7717/peerj.8419. PMID: 32071802; PMCID: PMC7008819. Abitew et al 2020 available here

Abitew-et-al-2020b(1,2)

Abitew DB, Yalew AW, Bezabih AM, Bazzano AN. Predictors of relapse of acute malnutrition following exit from community-based management program in Amhara region, Northwest Ethiopia: An unmatched case-control study. PLoS One. 2020 Apr 22;15(4):e0231524. doi: 10.1371/journal.pone.0231524. PMID: 32320426; PMCID: PMC7176369. Abitew et al 2020b available here

Adegok-et-al-2020(1,2)

Adegoke O, Arif S, Bahwere P, Harb J, Hug J, Jasper P, Mudzongo P, Nanama S, Olisenekwu G, Visram A. Incidence of severe acute malnutrition after treatment: A prospective matched cohort study in Sokoto, Nigeria. Matern Child Nutr. 2021 Jan;17(1):e13070. doi: 10.1111/mcn.13070. Epub 2020 Aug 5. PMID: 32761792; PMCID: PMC7729648. Adegok et al 2020 available here

Chang-et-al-2013

Chang CY, Trehan I, Wang RJ, Thakwalakwa C, Maleta K, Deitchler M, Manary MJ. Children successfully treated for moderate acute malnutrition remain at risk for malnutrition and death in the subsequent year after recovery. J Nutr. 2013 Feb;143(2):215-20. doi: 10.3945/jn.112.168047. Epub 2012 Dec 19. PMID: 23256140; PMCID: PMC3735907. Chang et al 2013 available here

Girma-et-al-2022(1,2,3,4,5,6,7,8)

Girma T, James PT, Abdissa A, Luo H, Getu Y, Fantaye Y, Sadler K, Bahwere P. Nutrition status and morbidity of Ethiopian children after recovery from severe acute malnutrition: Prospective matched cohort study. PLoS One. 2022 Mar 10;17(3):e0264719. doi: 10.1371/journal.pone.0264719. PMID: 35271590; PMCID: PMC8912152. Girma et al 2022 available here

King-et-al-2022

King S, D’Mello-Guyett L, Yakowenko E, Riems B, Gallandat K, Mama Chabi S, Mohamud FA, Ayoub K, Olad AH, Aliou B, Marshak A, Trehan I, Cumming O, Stobaugh H. A multi-country, prospective cohort study to measure rate and risk of relapse among children recovered from severe acute malnutrition in Mali, Somalia, and South Sudan: a study protocol. BMC Nutr. 2022 Aug 24;8(1):90. doi: 10.1186/s40795-022-00576-x. PMID: 36002905; PMCID: PMC9404649. King et al 2022 available here

Lambebo-et-al-2021

Lambebo A, Temiru D, Belachew T. Frequency of relapse for severe acute malnutrition and associated factors among under five children admitted to health facilities in Hadiya Zone, South Ethiopia. PLoS One. 2021 Mar 25;16(3):e0249232. doi: 10.1371/journal.pone.0249232. PMID: 33765081; PMCID: PMC7993841. Lambebo et al 2021 available here

Stobaugh-et-al-2018

Stobaugh HC, Rogers BL, Webb P, Rosenberg IH, Thakwalakwa C, Maleta KM, Trehan I, Manary MJ. Household-level factors associated with relapse following discharge from treatment for moderate acute malnutrition. Br J Nutr. 2018 May;119(9):1039-1046. doi: 10.1017/S0007114518000363. Epub 2018 Mar 5. PMID: 29502542. Stobaugh et al 2018 available here

Stobaugh-et-al-2019(1,2,3)

Stobaugh HC, Mayberry A, McGrath M, Bahwere P, Zagre NM, Manary MJ, Black R, Lelijveld N. Relapse after severe acute malnutrition: A systematic literature review and secondary data analysis. Matern Child Nutr. 2019 Apr;15(2):e12702. doi: 10.1111/mcn.12702. Epub 2018 Oct 18. PMID: 30246929; PMCID: PMC6587999. Stobaugh et al 2019 available here

Trehan-et-al-2015

Trehan I, Banerjee S, Murray E, Ryan KN, Thakwalakwa C, Maleta KM, Manary MJ. Extending supplementary feeding for children younger than 5 years with moderate acute malnutrition leads to lower relapse rates. J Pediatr Gastroenterol Nutr. 2015 Apr;60(4):544-9. doi: 10.1097/MPG.0000000000000639. PMID: 25419681; PMCID: PMC4380557. Trehan et al 2015 available here