*All 12 studies were at risk of performance bias because the participants/parents/clinical team were not masked to intervention. In all except Mazumder’s study (weightage 65.4%), the outcome assessors were also not masked to the intervention. However, mortality being a ‘hard’ outcome, we did not downgrade for either performance or outcome assessment bias. Six studies, Acharya, Boo, Cattaneo, Charpak, Eka Prawiti, and Worku, contributing to 26.3% weightage in the pooled analysis, are at unclear risk of allocation concealment. Four studies - Boo, Cattaneo, Suman, and Worku - were also at risk of attrition bias due to incomplete outcome data. But they together account for only 22.7% weightage in the pooled analysis. The risk of bias was therefore not downgraded to ‘serious’ risk. One study Mwendwa 2012 was at high risk of bias for random sequence generation and allocation concealment. It contributed to 3.5% weightage. The total number of neonates enrolled is quite large (~10 500) – therefore, the evidence was not downgraded for imprecision.

†All studies were at moderate or severe risk of bias as participants and outcome assessors were not masked to intervention and outcomes. Only in Mazumder et al the assessors were masked to the intervention. Though culture-positive sepsis is a ‘hard’ outcome, the largest study Mazumder 2019 that accounted for 91% of weightage defined sepsis based on WHO PSBI signs and not on culture positivity; five studies (4.7% weightage; Ali 2009, Eka Prawiti 2009, Kadam 2005, Kumbhojkar 2016, Suman 2008) did not define sepsis in their studies; another study (Rojas 2008; weightage 9.2%) defined it as both clinical and culture-positive sepsis, and only Boo 2007 defined it as culture-positive sepsis. Therefore, the risk of bias was downgraded to ‘serious’ risk. Allocation concealment was unclear in four studies (Charpak 1997, Ali 2009 Boo, 2007 Eka Pratiwi 2009) that together contribute to 4.8% weightage.

‡All studies were at high risk of outcome ascertainment bias as participants and outcome assessors were not masked to intervention and outcomes. However, weight gain is considered a ‘hard’ outcome. Therefore, we did not downgrade for the risk of bias. Seven studies (Acharya, Ali, Bier, Boo, Cattaneo, Gathwala, and Ramanathan accounting for 64% weightage) were at risk of allocation concealment bias. Therefore, the evidence was downgraded for ‘serious’ risk of bias.

§Substantial heterogeneity >50%.

¶All studies were at high risk of outcome ascertainment bias as participants and outcome assessors were not masked to intervention and outcomes. However, weight gain is considered a ‘hard’ outcome. Therefore, we did not downgrade for the risk of bias. Seven studies (Acharya, Ali, Bier, Boo, Cattaneo, Gathwala, and Ramanathan accounting for 64% weightage) were at risk of allocation concealment bias. Therefore, the evidence was downgraded for ‘serious’ risk of bias.

**All studies were at high risk of outcome ascertainment bias because the participants and outcome assessors were not masked to the intervention and the outcome was not a ‘hard’ outcome. Allocation concealment was unclear in six studies that accounted for 82% of weightage.

††95% CI overlap no effect (ie, CI includes RR of 1.0).

‡‡One study Charpak 1997 with moderate risk of bias (unclear allocation concealment; lack of blinding of participants/parents/clinical team and outcome assessors). The follow-up rate at 12–18 months was 80%. The characteristics of infants of KMC and conventional groups who completed follow-up were similar.

§§Single study.

MD, Mean difference; PMA, Postmenstrual age; RR, Risk ratio.