Description of the study

This study was a hospital-based prospective cohort study nested under the COAST-Nutrition trial (ISRCTN10829073, 06/06/2018 and PACTR202106635355751, Protocol Ref Makerere: #REC REF 2020 − 155). It was conducted between June 2020 and October 2022 in Uganda (Mbale Regional Referral Hospital, Soroti Regional Referral Hospital, Jinja Regional Referral Hospital, Masaka Regional Referral Hospital) and in Kenya (Kilifi County Hospital). Children were enrolled in the COAST-Nutrition trial if: aged 6 months to 12 years and hospitalized for severe pneumonia as defined by WHO (cough and or difficulty in breathing with chest indrawing and general danger signs) [18] ; had hypoxemia (pulse oximetry reading of SPO2 < 92% recorded in room air over 5 min; and parents or guardians consented to participate in the study. Children with severe malnutrition (MUAC < 11.5 cm, and/or the presence of bilateral oedema), known chronic lung disease, or congenital cardiac disease were excluded from the clinical trial. Children were randomized to nutritional supplementation with Ready-to-Use Therapeutic Foods (RUTF) in addition to usual diet for 56 days compared to usual diet al.one (control). Details of the main clinical trial under which this study was nested are published in the trial protocol [19].

Study and trial specific assessmentsParticipant recruitment and follow up

At the point of hospital admission, eligible children with suspected severe pneumonia and hypoxaemia (oxygen saturations on pulse oximetry of < 92%) were screened and consecutively enrolled. Parental consent was obtained for inclusion in the trial and additional anthropometric measurements. All study participants were followed up until death or study completion (Day 180), whichever occurred first. Full details of the trial protocol including the anthropometric sub-studies have been published [19]. In this sub-study we only included children enrolled into the control arm (received only their usual diet, without supplemental feeding with ready to use therapeutic food, in addition to their pneumonia treatment) and were aged 6–59 months.

Assessment of independent variables

Participants’ history (symptoms at presentation to hospital and associated factors) was obtained from caregivers and entered into a questionnaire. Clinical examination was done by medical officers and the findings as well as the results of the laboratory investigations at admission to hospital were entered in a questionnaire.

Anthropometric measures

All children enrolled into the study had baseline measurements of mid-upper arm circumference (MUAC) measured with a recommended non-stretchable tape to the nearest millimeter. Triceps skinfold thickness (TST) was measured with a special caliper (Harpenden ®) by trained research staff to the nearest 0.2 mm. Two measurements were taken and the average was recorded [19]. If there were significant differences between the two measurements taken, another research assistant independently took his/her own measurements and the average was recorded. Research assistants at all sites were trained on taking anthropometric measurements prior to starting data collection and periodically for the duration of the study.

Estimation of body composition

Body composition is most accurately assessed by invasive and clinically impractical procedures such as densitometry and isotope dilution for total body water [20]. Some less invasive methods use the two-compartment model to describe overall body composition by dividing the body into two compartments: fat mass (FM) and fat free mass (FFM) or lean mass which includes muscle, bone and body organs, without giving information on distribution [20]. One of the relatively simple techniques utilizes anthropometric indices incorporating MUAC and triceps skinfold thickness (TST) to measure adipose tissue distribution and estimate overall body composition using indices like arm fat area (AFA), arm muscle area (AMA) and arm muscle circumference (AMC) [21]. Arm fat area and arm muscle area calculations are based on assumptions that: (i) the arm is cylindrical; (ii) subcutaneous fat is evenly distributed around a circular core of muscle; and (iii) Triceps skinfold thickness (TST) reflects the fat components of the arm and represents twice the thickness of subcutaneous fat in the arm [21].

When the anthropometric indices were evaluated for validity among healthy and sick children (with cystic fibrosis) using dual x-ray absorptiometry and four-component model to provide reference values for arm and whole-body fat mass (FM) and fat-free mass (FFM), Chomtho et al. found that arm fat area (AFA), MUAC, and TST correlated strongly with arm FM (r = 0.84–0.92) and total FM (r = 0.78–0.92). However, AMA and MUAC are weakly correlated with fat free mass indices [17]. Rolland-Cachera et al. validated the arm muscle area, arm fat area, and arm circumference against magnetic resonance imaging (MRI) and found that these were indeed accurate for assessment of body composition [22]. We retrospectively calculated Arm muscle circumference (AMC), arm muscle area (AMA) and arm fat area (AFA) by the equations proposed by Rolland-Cachera. AMC, AFA and AMA were estimated from MUAC and TST. The AMC was calculated with the formula AMC = MUAC – π TST where π = 22/7. The AFA is calculated from the formula MUAC × TST/2. The AMA was calculated with the formula; AMA = (MUAC)²/4π – MUAC × TST/2 [22]. This is illustrated in Fig. 1.

Fig. 1

Transverse section of the arm showing assumptions of fat and muscle area. d1represents arm muscle area estimate (fat free mass), while d2 representsthe total arm area estimate. AFA = MUAC × TST/2, while AMA = (MUAC)²/4π – MUAC × TST/2

Sample size estimation

This was a sub-analysis of participants aged 6–59 months who were randomized to the control (usual diet) arm of the COAST-Nutrition trial (Fig. 2). We used a convenient sample for this study which was based on the available participants from the main trial in Uganda and Kenya. With an Area Under the Curve (AUC) of 0.8, mortality of 14% [1], a sample size of 369 children gave a confidence interval width of 0.15 for a confidence level of 0.95 [23].

Fig. 2

Study flow reflecting participants included in this sub-study of the main trial. A total of 369 participants were included in this study

Data analysis

The primary outcome was survival at day 180 of follow up (expressed as a proportion). The independent variables included MUAC, arm muscle circumference (AMC), arm muscle area (AMA), arm fat area (AFA) as well as age, gender and history (symptoms at presentation), clinical examination findings and laboratory results at admission to hospital (baseline). Receiver operator characteristic (ROC) curves for MUAC, AMC, AMA, and AFA on admission were generated using STATA version MP16 (StataCorp LLC, College Station, Texas, USA), and examined with survival at day 180 of follow-up as the outcome. The fat and muscle mass indices most predictive of the outcome were expected to have a greater AUC than other indices. The AUC for AFA was compared to that of AMA and AMC using the Chi-square test.

The AUC for AMC, AMA, and AFA were also compared to that of MUAC, currently the best predictor of mortality [24], using the Chi-square test. We used the Cox survival analysis model to generate hazard ratios for AMC, AMA, and AFA to that of MUAC, while controlling for baseline risk factors.

All analyses were adjusted for age in months, gender, breastfeeding status and respiratory rate at baseline (admission to hospital). These were selected based on biological plausibility using available literature, as well as having sufficient numbers of outcomes for comparison.