Time to diagnosis of acute kidney injury and its predictors among outborn neonates in India: a competing-risk analysis of tele-ICU network data (2022–2024) Authors: Sitarah Mathias, Ryan Brewster, Eva Robinson, Carl Britto, Michael Monuteaux, Prashantha YN, Dileep Unnikrishnan, Geetanjali Srivastava, Ankana Daga, Lakshmi Ganapathi Published: 2026 ABSTRACT SummaryBackgroundThere is limited data on neonatal acute kidney injury (AKI) from resource-constrained settings. In this study, we describe time to AKI diagnosis and its predictors among neonates born outside the treating hospital (outborn) in India, accounting for competing events like death and discharge against medical advice (DAMA).MethodsThis multicentre retrospective study leveraged data from a tele-NICU network across 27 sites in India from May 2022–June 2024. It included neonates 28 days or younger at admission with two or more serum creatinine values measured within a seven-day span, beginning after 48 h of life. AKI was determined using neonatal modified Kidney Disease: Improving Global Outcomes (KDIGO) criteria. Primary outcome was time-to-AKI diagnosis. Cumulative incidence function was used to estimate AKI risk over time and Cox proportional hazards model was used to identify predictors amid competing events. Secondary outcomes like death, DAMA and length of hospitalisation were described.FindingsAmong 570 neonates, sex was recorded for 506; of these, 411 (81.2%) were male and 95 (18.8%) were female. The mean gestational age was 35.4 weeks (SD 3.1; range 24.3–42.7 weeks). Median post-natal age at admission was 2 days (IQR 1–5). AKI occurred in 170 (29.8%) neonates. Median time-to-AKI diagnosis was 2 days (IQR 1–5) post-admission; five-day cumulative incidence was 0.23 (95% CI: 0.20–0.27). Significant predictors of early AKI included dehydration (HR 3.63, 95% CI: 1.67–7.92), mechanical ventilation (HR 1.84, 95% CI: 1.32–2.56) and vasopressor use (HR 1.33, 95% CI: 1.05–1.69). AKI was associated with higher DAMA (34.7% vs. 25.2%, p = 0.02) and Death (13.5% vs. 5.3%, p < 0.01).InterpretationOutborn neonates have rapid time-to-AKI diagnosis post-admission, highlighting a critical high-risk window for kidney-protective interventions. Policy changes to strengthen medical transport and referral systems may additionally reduce AKI risk among outborn neonates.FundingFred Lovejoy Resident Research and Education Award (Boston Children’s Hospital). Research in contextEvidence before this studyWe searched Pubmed and Google Scholar for studies examining risk factors and outcomes of neonatal acute kidney injury (AKI) from inception till October 2024. Search terms included combinations of “neonates” AND “acute kidney injury” AND (“risk factors” OR “predictors” OR “outcomes”). We did not include language filters but included observational studies, systematic reviews and meta-analyses. Existing studies from low-income and middle-income countries (LMIC) report widely varying incidence rates of neonatal AKI and demonstrated associations with poor outcomes (increased death, length of hospitalization and costs). However, most available data are derived from inborn neonates from academic or tertiary centers, and do not address risk factors and outcomes among outborn neonates (born at home or outside the hospital). We did not identify studies evaluating time-to-AKI diagnosis among outborn neonates in LMIC like India. Furthermore, discharge against medical advice (DAMA) is infrequently reported, and most studies do not account for mortality or DAMA as competing risk factors when studying AKI and its outcomes. Therefore, important gaps remain in understanding burden, timing, predictors and outcomes of AKI among outborn neonates.Added value of this studyThis study addresses important evidence gaps regarding AKI by incorporating competing-risk analyses that account for mortality and DAMA among outborn neonates. It provides an assessment of AKI occurrence and timing in resource-limited settings including the identification of a high-risk period for potential interventions. The findings highlight a high AKI burden, affecting 170/570 (29.8%) neonates, with a median time to AKI diagnosis of 2 days (IQR 1–5) post-admission, indicating an early high-risk window. The study also extends current understanding of potentially modifiable contributors to early AKI, which included dehydration (HR 3.63, 95% CI: 1.67–7.92), mechanical ventilation (HR 1.84, 95% CI: 1.32–2.56) and vasopressor use (HR 1.33, 95% CI: 1.05–1.69), highlighting the possible influence of pre-admission and transport-related factors on neonatal kidney injury.Implications of all the available evidenceTaking together with existing evidence, our findings underscore the urgency for early recognition and prevention of neonatal AKI. The early time-to-AKI diagnosis among outborn infants raises concern for pre-admission onset of renal injury and identifies a high-risk window for the implementation of “nephroprotective bundles” (including risk stratification, screening protocols and fluid balance optimization) to improve detection and reduce AKI risk. Our data also highlights critical data gaps pertaining to pre-admission factors, the potential role of transport delay in increasing the risk of AKI in outborn neonates, as well as the influence of socio-economic factors in determining outcomes (DAMA). In addition to the implementation of nephrotoxic bundles at admission for outborn neonates, strengthening referral pathways, improving neonatal transport systems, and enhancing maternal and neonatal healthcare delivery may help reduce preventable kidney injury and improve neonatal outcomes in resource-limited settings. Research in context Evidence before this study We searched Pubmed and Google Scholar for studies examining risk factors and outcomes of neonatal acute kidney injury (AKI) from inception till October 2024. Search terms included combinations of “neonates” AND “acute kidney injury” AND (“risk factors” OR “predictors” OR “outcomes”). We did not include language filters but included observational studies, systematic reviews and meta-analyses. Existing studies from low-income and middle-income countries (LMIC) report widely varying incidence rates of neonatal AKI and demonstrated associations with poor outcomes (increased death, length of hospitalization and costs). However, most available data are derived from inborn neonates from academic or tertiary centers, and do not address risk factors and outcomes among outborn neonates (born at home or outside the hospital). We did not identify studies evaluating time-to-AKI diagnosis among outborn neonates in LMIC like India. Furthermore, discharge against medical advice (DAMA) is infrequently reported, and most studies do not account for mortality or DAMA as competing risk factors when studying AKI and its outcomes. Therefore, important gaps remain in understanding burden, timing, predictors and outcomes of AKI among outborn neonates. Added value of this study This study addresses important evidence gaps regarding AKI by incorporating competing-risk analyses that account for mortality and DAMA among outborn neonates. It provides an assessment of AKI occurrence and timing in resource-limited settings including the identification of a high-risk period for potential interventions. The findings highlight a high AKI burden, affecting 170/570 (29.8%) neonates, with a median time to AKI diagnosis of 2 days (IQR 1–5) post-admission, indicating an early high-risk window. The study also extends current understanding of potentially modifiable contributors to early AKI, which included dehydration (HR 3.63, 95% CI: 1.67–7.92), mechanical ventilation (HR 1.84, 95% CI: 1.32–2.56) and vasopressor use (HR 1.33, 95% CI: 1.05–1.69), highlighting the possible influence of pre-admission and transport-related factors on neonatal kidney injury. Implications of all the available evidence Taking together with existing evidence, our findings underscore the urgency for early recognition and prevention of neonatal AKI. The early time-to-AKI diagnosis among outborn infants raises concern for pre-admission onset of renal injury and identifies a high-risk window for the implementation of “nephroprotective bundles” (including risk stratification, screening protocols and fluid balance optimization) to improve detection and reduce AKI risk. Our data also highlights critical data gaps pertaining to pre-admission factors, the potential role of transport delay in increasing the risk of AKI in outborn neonates, as well as the influence of socio-economic factors in determining outcomes (DAMA). In addition to the implementation of nephrotoxic bundles at admission for outborn neonates, strengthening referral pathways, improving neonatal transport systems, and enhancing maternal and neonatal healthcare delivery may help reduce preventable kidney injury and improve neonatal outcomes in resource-limited settings. Introduction Neonatal acute kidney injury (AKI) is associated with mortality,1,2 prolonged hospitalisations1 and increased healthcare expenditure.3 Several studies have linked neonatal intensive care unit (NICU) hospitalisations and neonatal AKI with an increased lifetime risk of chronic kidney disease (CKD).4, 5, 6 Most current epidemiologic data on neonatal AKI originate from high-income countries (HICs) and tertiary centres with better access to diagnostics, nephrology expertise and therapeutic resources.1,2 Conversely, data from low-income and middle-income countries (LMICs) are heterogenous.7,8 LMICs have unique features including high outborn (born outside the hospital/at home) rates, under-developed medical transport infrastructure and high rates of competing outcomes (early mortality and discharge against medical advice).9, 10, 11 Resource constraints in these settings can also limit testing frequency and management strategies, ultimately influencing reported diagnostic rates and outcomes including mortality.12 Diagnosis of neonatal AKI is challenging because current criteria rely on derangements in serum creatinine and urine output which lag the onset of kidney injury, and require interpretation within the contexts of gestational age (GA) and illness severity.13,14 These challenges are amplified in settings with limited diagnostic capability and therapeutic options.12 Increased paediatric CKD progression and mortality demonstrated in LMICs15 emphasise the need for prevention, early diagnosis and timely management of neonatal AKI. Tele-NICU networks provide NICU services to remote, resource-limited settings, and enable large multicentric epidemiological studies through relatively standardised data collection across geographically and culturally diverse sites.16 While the provision of clinical advice may be relatively uniform by virtue of a consistent remote medical team, substantial variability in baseline resources, implementation of these services and protocol adherence result in heterogeneity of clinical care at peripheral sites. In this study, we leveraged data from a multicentre tele-NICU network spanning 27 NICUs in India. The objective was to characterise the incidence, timing, predictors and outcomes of AKI among outborn neonates to inform future risk-reduction interventions. Methods This retrospective study was conducted between June 2022 to May 2024, and leveraged data from 27 sites (Supplementary Fig. S1 and Table S1) within an Indian Tele-NICU consortium (Cloudphysician.AI). They utilised indigenous technology and proprietary software to provide NICU expertise through a “hub-and-spoke” model. The Cloudphysician neonatology team (hub) provided continuous virtual clinical support to their partner sites (spokes), with daily interactions documented within their electronic medical records (EMR) system, thus providing a unified database for data collection. A detailed description of their clinical operations and implementation process can be found elsewhere.16,17 Partner sites included public or private institutions in urban or rural settings with varying baseline resources. Depending on population size and infrastructure, partner sites are in tier-1 (metropolitan), tier-2 (developed) or tier-3 (developing) cities, a classification that indirectly signifies available resources18 (Supplementary Table S1). All admitted neonates 28 days or younger within the Cloudphysician.AI network was included if they had two or more serum creatinine values within a seven-day span, both drawn after 48 h of life. Neonates with known congenital renal anomalies, obstructive renal pathology, or who were discharged, transferred or died within 48 h of life, were excluded (Fig. 1). Decisions surrounding patient eligibility occurred through a consensus process involving authors SM, AD and LG to minimise bias and maintain consistency. A minimum sample size of 323 neonates was calculated for our study (Supplementary methods). Neonates were assigned to the AKI group they had at least on episode of AKI (identified through diagnostic code or using the neonatal modified KDIGO (Kidney Disease: Improving Global Outcomes) criteria13 (Supplementary Table S2) for AKI diagnosis). Available data were insufficient to reliably record AKI stages or recurrent episodes.Fig. 1Patient eligibility assessment and inclusion. Legend: This figure denotes patient eligibility assessment for inclusion, numbers excluded with reasons for exclusion. This follows the STROBE guidelines. Patient eligibility assessment and inclusion. Legend: This figure denotes patient eligibility assessment for inclusion, numbers excluded with reasons for exclusion. This follows the STROBE guidelines. Anonymised demographic and clinical data (comorbidities, medication and laboratory data) were extracted from data dashboards within the EMR system. Initial data collection occurred through a keyword-based text query, followed by manual verification of the full dataset by a second reviewer. Explicitly mentioned risk factor diagnoses and objective data were collected from admission till AKI diagnosis in the AKI group, and from admission to discharge in the non-AKI group. Retrospective clinical interpretations were not undertaken and discrepancies between both rounds of data collection were resolved using clearly documented risk factor diagnoses or objective data. In its absence, the most conservative approach was undertaken, and the variable was recorded as missing. Prenatal and birth records including resuscitation details were unavailable. Definitions of collected clinical data are included in the Supplementary section (Supplementary Tables S3 and S4). The primary outcome was time-to-AKI diagnosis following admission. Secondary outcomes included length of hospitalisation (LOH), in-hospital mortality (Death), Discharge against medical advice (DAMA), discharge to home/lower level of care (recovery) or transfer to similar or higher level of care. All disposition outcomes were discretely recorded in the EMR system in real-time. This manuscript follows the STROBE reporting guidelines.19 Statistical analysis Descriptive statistics for baseline data were calculated using proportions (categorical variables) and median with inter-quartile range (IQR) (continuous variables). Fisher’s exact test and Wilcoxon rank sum test were used to assess associations between AKI and categorical and continuous variables, respectively. The AKI incidence (percentage) was calculated with 95% confidence intervals (CI). Death, DAMA, recovery, and transfer to higher level of care were summarised (proportions) and compared by AKI status using Fisher’s exact test. Median LOH were compared between AKI and non-AKI groups using Wilcoxon rank sum test and repeated after excluding patients who experienced death or DAMA. Median time-to-AKI from admission was calculated, and cumulative incidence function (CIF) was used to estimate the risk of AKI, death and DAMA over time. A cause-specific cox proportional hazard model was performed to evaluate predictors of time-to-AKI, treating death and DAMA as competing risks. Patients discharged from the hospital without experiencing AKI, death, or DAMA were treated as censored. The distinction between competing risks and censoring was based on each outcome’s implications for continued AKI risk and our ability to observe it. DAMA occurs before treatment completion, is frequently precipitated by high out-of-pocket healthcare costs and is strongly associated with post-discharge mortality.9,20,21 As post-discharge clinical data were not available, patients experiencing DAMA were considered to have discontinued treatment without eliminating AKI risk, while simultaneously becoming unobservable for subsequent AKI episodes. Similarly, in-hospital mortality before AKI precludes subsequent AKI without eliminating its underlying risk. Therefore, both DAMA and death were considered competing events. In contrast, patients routinely discharged without AKI were censored because they were considered clinically stable and without active AKI risk by the treating team. Although AKI could theoretically develop after discharge, this was considered clinically less likely and not observable. Patients transferred to similar or higher levels of care before developing AKI were also similarly censored assuming continued treatment for underlying AKI risk factors at the receiving facility. Reasons for transfer (clinical, logistic, social etc) were not captured in our dataset, precluding the ascertainment of the level of AKI risk at transfer. A cluster-robust sandwich variance estimator at the hospital level was used to account for potential within-hospital correlation.22 The proportional hazards assumption was tested using Schoenfeld residuals. Covariates included prematurity, Low Birth Weight (LBW), presence of sepsis, heart disease, mechanical ventilation, dehydration, vasopressor use, nephrotoxin exposure, necrotizing enterocolitis (NEC), and hypoxic ischemic encephalopathy (HIE). Due to high missingness in the reporting of LBW (20.2%), a multivariate imputation by chained equations (MICE) method using ten imputations was performed. A complete case analysis was run as a sensitivity analysis. The relative variance increase (RVI) and fraction of missing information (FMI) are reported as imputation diagnostics. Variance inflation factors (VIF) were also calculated to test for possible collinearity between covariates for both the complete case analysis and the multiply imputed model, and the threshold for collinearity considered was >5.23 A second competing risk analysis for LOH was performed using the same covariates mentioned above, in addition to AKI, while considering death and DAMA as competing risks. A post-hoc exploratory analysis tested for the possible non-linear effect of GA on AKI by including a quadratic term in the model. An alpha level of 0.05 was used for all tests. All calculations were performed using R version 4.5.0.24 Ethics statement This study met criteria for non-human participant research and qualified for an informed consent waiver. It was approved by the Boston Children’s Hospital Institutional Review Board (IRB No: IRB-P00045713) and Cachar Cancer Hospital on behalf of all collaborating local Indian sites (IRB No:ECR/925/InstlASI-2017/RR-21, ECIN EWll N ST/2021). Role of the funding source Funders were not involved in the design, planning or conduction of the study, data analysis or manuscript preparation. Results Among 4405 neonates admitted over two years across 27 sites (Supplementary Table S1), 662 met the laboratory inclusion criteria. Ninety neonates were excluded for missing/ineligible age, one had posterior urethral valve, and one had bilateral ureteric compression secondary to an abdominal tumour (Fig. 1). All 570 included neonates were outborn, and most of them (n = 522, 91.6%) presented to a site in a tier-3 city (Supplementary Table S1). The median post-natal age at admission was 2 days (IQR 1–5). Majority of the neonates were male (n = 411, 81.2%) and premature (n = 378, 66.5%). The median GA was 36.1 weeks (IQR 34.0–37.3; range: 24.3–42.7 weeks), with 57.9% (n = 330) between 30 and 36 completed weeks. Among 440 neonates with birth weight data, median birth weight was 2.5 kg (IQR 1.9–2.9). Over half the neonates were mechanically ventilated (n = 352, 62.0%), required vasopressors (n = 335, 58.8%) or were diagnosed with sepsis (n = 357, 63.1%). HIE was reported in 246 neonates (43.3%). Nephrotoxin exposure was recorded in 411 neonates (72.1%) with 206 neonates (36.1%) receiving two or more nephrotoxins during their hospitalisation (Table 1).Table 1Baseline demographics and clinical characteristics of 570 outborn neonates admitted to 27 Tele-NICU sites in India, stratified by AKI diagnosis during NICU stay.Patient demographicsCharacteristicOverall N = 570Neonatal AKIp-valueNo N = 400Yes N = 170Demographic information Sex, n (%)0.80 Female95 (18.8)70 (19.2)25 (17.7) Male411 (81.2)295 (80.8)116 (82.3) Unknown643529 Postnatal age at admission (days) Median (IQR)2 (1.0–5.0)2 (1.0–5.0)3 (1.0–6.0)0.03 Prematuritya, n (%)378 (66.5)274 (68.5)104 (61.9)0.14 Unknown202Gestational age (weeks) Mean (SD)35.4 (3.1)35.3 (3.2)35.9 (2.8) Range (min–max)24.3–42.724.3–42.726.0–42.7 Median (IQR)36.1 (34.0–37.3)36.1 (33.6–37.1)36.3 (34.6–38.0)0.07Gestational age subgroups, n (%)0.07 <30 weeks’ gestation44 (7.7)37 (9.3)7 (4.1) 30–36 and 6/7 weeks’ gestation330 (57.9)234 (58.5)96 (56.5) ≥37 weeks’ gestation186 (32.6)124 (31.0)62 (36.5) Unknown1055Birth weight (kg) Median (IQR)2.5 (1.9–2.9)2.4 (1.8–2.9)2.6 (2.1–3.0)0.03 Unknown1308050Low birth weightb, n (%)220 (48.4)169 (51.2)51 (40.8)0.06 Unknown1157045Clinical characteristics Respiratory supportc, n (%)487 (85.7)341 (85.7)146 (85.9)>0.90 Unknown220 Invasive mechanical ventilation, n (%)352 (62.0)225 (56.5)127 (74.7)<0.01 Unknown220 Hyperbilirubinemia, n (%)155 (27.5)121 (30.8)34 (20.0)0.01 Unknown770 Necrotizing Enterocolitis, n (%)33 (5.9)26 (6.6)7 (4.1)0.30 Unknown770 Hypoxic Ischemic Encephalopathy, n (%)246 (43.3)162 (40.7)84 (49.4)0.07 Unknown220 Surgeryd, n (%)14 (2.5)14 (3.6)0 (0.0)0.01 Unknown10100 Central linee, n (%)38 (6.7)30 (7.5)8 (4.7)0.30 Sepsisf, n (%)357 (63.1)249 (62.9)108 (63.5)>0.90 Unknown440 Dehydration, n (%)52 (9.2)29 (7.3)23 (13.5)0.03 Unknown330 Heart disease, n (%)32 (5.6)26 (6.5)6 (3.5)0.20 Unknown330 Vasopressor use, n (%)335 (58.8)217 (54.3)118 (69.4)<0.01Nephrotoxin medication exposure (days) Median (IQR)2 (0–6)3 (1–8)1 (0–4)<0.01Nephrotoxic drugs combinations, n (%)<0.01 Single drug205 (36.0)153 (38.3)52 (30.6) 2 or more drugs206 (36.1)160 (40.0)46 (27.1) No drugs159 (27.9)87 (21.8)72 (42.4)Antimicrobial use, n (%)413 (72.5)314 (78.5)99 (58.2)<0.01Caffeine use, n (%)185 (32.5)147 (36.8)38 (22.4)<0.01Furosemide use, n (%)97 (17.0)60 (15.0)37 (21.8)0.05IQR: Inter-quartile range; SD: Standard deviation.(a)Prematurity: <37 weeks’ gestation.(b)Low birth weight: <2.5 kg.(c)Respiratory support includes invasive and non-invasive support.(d)Surgery: major cardiac or abdominal surgery.(e)Central line: includes umbilical venous or arterial catheter or peripherally inserted central catheter.(f)Sepsis: diagnosed per clinical criteria or positive blood cultures. Table 1Baseline demographics and clinical characteristics of 570 outborn neonates admitted to 27 Tele-NICU sites in India, stratified by AKI diagnosis during NICU stay.Patient demographicsCharacteristicOverall N = 570Neonatal AKIp-valueNo N = 400Yes N = 170Demographic information Sex, n (%)0.80 Female95 (18.8)70 (19.2)25 (17.7) Male411 (81.2)295 (80.8)116 (82.3) Unknown643529 Postnatal age at admission (days) Median (IQR)2 (1.0–5.0)2 (1.0–5.0)3 (1.0–6.0)0.03 Prematuritya, n (%)378 (66.5)274 (68.5)104 (61.9)0.14 Unknown202Gestational age (weeks) Mean (SD)35.4 (3.1)35.3 (3.2)35.9 (2.8) Range (min–max)24.3–42.724.3–42.726.0–42.7 Median (IQR)36.1 (34.0–37.3)36.1 (33.6–37.1)36.3 (34.6–38.0)0.07Gestational age subgroups, n (%)0.07 <30 weeks’ gestation44 (7.7)37 (9.3)7 (4.1) 30–36 and 6/7 weeks’ gestation330 (57.9)234 (58.5)96 (56.5) ≥37 weeks’ gestation186 (32.6)124 (31.0)62 (36.5) Unknown1055Birth weight (kg) Median (IQR)2.5 (1.9–2.9)2.4 (1.8–2.9)2.6 (2.1–3.0)0.03 Unknown1308050Low birth weightb, n (%)220 (48.4)169 (51.2)51 (40.8)0.06 Unknown1157045Clinical characteristics Respiratory supportc, n (%)487 (85.7)341 (85.7)146 (85.9)>0.90 Unknown220 Invasive mechanical ventilation, n (%)352 (62.0)225 (56.5)127 (74.7)<0.01 Unknown220 Hyperbilirubinemia, n (%)155 (27.5)121 (30.8)34 (20.0)0.01 Unknown770 Necrotizing Enterocolitis, n (%)33 (5.9)26 (6.6)7 (4.1)0.30 Unknown770 Hypoxic Ischemic Encephalopathy, n (%)246 (43.3)162 (40.7)84 (49.4)0.07 Unknown220 Surgeryd, n (%)14 (2.5)14 (3.6)0 (0.0)0.01 Unknown10100 Central linee, n (%)38 (6.7)30 (7.5)8 (4.7)0.30 Sepsisf, n (%)357 (63.1)249 (62.9)108 (63.5)>0.90 Unknown440 Dehydration, n (%)52 (9.2)29 (7.3)23 (13.5)0.03 Unknown330 Heart disease, n (%)32 (5.6)26 (6.5)6 (3.5)0.20 Unknown330 Vasopressor use, n (%)335 (58.8)217 (54.3)118 (69.4)<0.01Nephrotoxin medication exposure (days) Median (IQR)2 (0–6)3 (1–8)1 (0–4)<0.01Nephrotoxic drugs combinations, n (%)<0.01 Single drug205 (36.0)153 (38.3)52 (30.6) 2 or more drugs206 (36.1)160 (40.0)46 (27.1) No drugs159 (27.9)87 (21.8)72 (42.4)Antimicrobial use, n (%)413 (72.5)314 (78.5)99 (58.2)<0.01Caffeine use, n (%)185 (32.5)147 (36.8)38 (22.4)<0.01Furosemide use, n (%)97 (17.0)60 (15.0)37 (21.8)0.05IQR: Inter-quartile range; SD: Standard deviation.(a)Prematurity: <37 weeks’ gestation.(b)Low birth weight: <2.5 kg.(c)Respiratory support includes invasive and non-invasive support.(d)Surgery: major cardiac or abdominal surgery.(e)Central line: includes umbilical venous or arterial catheter or peripherally inserted central catheter.(f)Sepsis: diagnosed per clinical criteria or positive blood cultures. Baseline demographics and clinical characteristics of 570 outborn neonates admitted to 27 Tele-NICU sites in India, stratified by AKI diagnosis during NICU stay. IQR: Inter-quartile range; SD: Standard deviation. Prematurity: <37 weeks’ gestation. Low birth weight: <2.5 kg. Respiratory support includes invasive and non-invasive support. Surgery: major cardiac or abdominal surgery. Central line: includes umbilical venous or arterial catheter or peripherally inserted central catheter. Sepsis: diagnosed per clinical criteria or positive blood cultures. Males accounted for 81.2% (n = 411) of our dataset with only 16.7% (n = 95) female. Male neonates had a higher median birth weight and higher proportion receiving vasopressors compared to female neonates. Sex-disaggregated demographic and clinical characteristics are reported in the Supplementary section (Supplementary Table S5). AKI was diagnosed in 29.8% (n = 170) of all neonates with 56.5% between 30 and 36 weeks GA. Five neonates underwent peritoneal dialysis. AKI incidence did not vary between males and females (28.2% vs. 26.3% respectively, p = 0.81). Neonates diagnosed with AKI had a higher median post-natal age at admission [3 days (IQR 1–6) vs. 2 days (IQR 1–5), p = 0.03], and a higher median birth weight [2.6 kg (IQR 2.1–3.0) vs. 2.4 kg (IQR 1.8–2.9); p = 0.03] compared to the non-AKI group. There was no statistically significant difference in AKI incidence across the GA subgroups (Table 1). Compared to the non-AKI group, neonates diagnosed with AKI required mechanical ventilation (74.7% vs. 56.5%, p < 0.01) and vasopressors (69.4% vs. 54.3%, p < 0.01), were diagnosed with dehydration (13.5% vs. 7.3%, p = 0.03), or had a significantly lower proportion of hyperbilirubinemia (20.0% vs. 30.8%, p = 0.01), nephrotoxin exposure (57.6% vs. 78.2%, p < 0.01) and caffeine use (22.4% vs. 36.8%, p < 0.01). (Table 1). Median time-to-AKI diagnosis from admission was 2 days (IQR 1–4) with 75% of AKI occurring within the first five days of admission (cumulative incidence at 5 days–0.23 (95% CI: 0.20–0.27); Fig. 1, Supplementary Table S5). Cumulative incidence of AKI, Death and DAMA are shown in Fig. 2.Fig. 2Cumulative incidence of acute kidney injury, death and discharge against medical advice over time Legend: Cumulative incidence of acute kidney injury (in blue), death (in red) and discharge against medical advice (in green) over time (in days) from admission. Cumulative incidence of acute kidney injury, death and discharge against medical advice over time Legend: Cumulative incidence of acute kidney injury (in blue), death (in red) and discharge against medical advice (in green) over time (in days) from admission. In the Complete Case Analysis, the proportional hazard assumption of the Cox model was not rejected using a Schoenfeld residual test (p = 0.17). The MICE procedure had an average RIV of 0.02 and the largest FMI was 0.13. The hazard ratio (HR) estimates were comparable between the univariate and multivariate model without imputation (Table 2, Supplementary Fig. S2). Mechanical ventilation (HR 1.84, 95% CI: 1.32–2.56), dehydration (HR 3.63, 95% CI: 1.67–7.92) and vasopressor use (HR 1.33, 95% CI: 1.05–1.69 using the MICE procedure) were significant predictors of developing AKI early. Vasopressor use was not significant with the Complete Case Analysis (HR 1.30, 95% CI: 0.91–1.85) however the variance inflation factors (VIF) for both the Complete Case Analysis (3.2) and MICE procedure (2.3) were low (Table 2). Single nephrotoxin (HR 0.53, 95% CI: 0.39–0.70; p < 0.01) and multiple nephrotoxin exposures (HR 0.40, 95% CI: 0.30–0.55; p < 0.01) were associated with a lower hazard of developing AKI over the study period. The post-hoc evaluation for non-linear effect of GA on AKI was not statistically significant (HR 0.99, 95% CI: 0.98–1.01).Table 2Competing risk analysis for developing AKI over time with covariates.CovariatesUnivariate analysisMultivariate analysis (n = 441)Multivariate analysis- with MICE (n = 570)Coefficient (95% CI)p-valueCoefficient (95% CI)p-valueAdj VIFCoefficient (95% CI)p-valueAdj VIFGestational Age (n = 374)1.09 (1.04, 1.14)<0.011.07 (1.00, 1.15)0.052.51.04 (0.98, 1.11)0.193.4Low Birth Weight (n = 220)0.60 (0.32, 1.10)0.100.92 (0.64, 1.34)0.684.20.87 (0.58, 1.31)0.503.0Mechanical Ventilation (ref = none; n = 352)1.87 (1.07, 3.27)0.031.82 (1.25, 2.65)<0.012.91.84 (1.32, 2.56)<0.012.4Necrotizing Enterocolitis (ref = none; n = 33)0.56 (0.24, 1.29)0.170.53 (0.23, 1.24)0.142.00.68 (0.27, 1.70)0.412.3Hypoxic-Ischemic Encephalopathy (ref = none; n = 246)1.41 (0.84, 2.37)0.191.19 (0.71, 1.98)0.514.41.18 (0.87, 1.61)0.292.6Sepsis (ref = none; n = 357)1.00 (0.72, 1.39)1.001.11 (0.87, 1.43)0.392.51.14 (0.90, 1.45)0.281.8Dehydration (ref = none; n = 52)2.36 (1.14, 4.89)0.023.84 (1.34, 10.99)0.014.93.63 (1.66, 7.92)<0.012.8Single nephrotoxic medication (ref = none; n = 205)0.50 (0.32, 0.77)<0.010.46 (0.28, 0.75)<0.013.10.53 (0.39, 0.7)<0.011.9≥2 nephrotoxic medication (ref = none; n = 206)0.39 (0.24, 0.65)<0.010.39 (0.26, 0.59)<0.010.40 (0.30, 0.55)<0.01Heart Disease (ref = none; n = 32)0.55 (0.22, 1.36)0.200.49 (0.21, 1.13)0.093.30.63 (0.25, 1.58)0.323.2Vasopressor exposure (ref = none; n = 335)1.70 (1.10, 2.62)0.021.30 (0.91, 1.85)0.153.21.33 (1.05, 1.69)0.022.3Covariate reference values specified and event counts noted as “n”; Hazard ratio (HR) estimates for univariate and multivariate analyses for the complete case analysis as well as imputation model estimates are provided. In this analysis, Death and DAMA were considered competing events for AKI; Adj VIF = Average adjusted variance inflation factors. Table 2Competing risk analysis for developing AKI over time with covariates.CovariatesUnivariate analysisMultivariate analysis (n = 441)Multivariate analysis- with MICE (n = 570)Coefficient (95% CI)p-valueCoefficient (95% CI)p-valueAdj VIFCoefficient (95% CI)p-valueAdj VIFGestational Age (n = 374)1.09 (1.04, 1.14)<0.011.07 (1.00, 1.15)0.052.51.04 (0.98, 1.11)0.193.4Low Birth Weight (n = 220)0.60 (0.32, 1.10)0.100.92 (0.64, 1.34)0.684.20.87 (0.58, 1.31)0.503.0Mechanical Ventilation (ref = none; n = 352)1.87 (1.07, 3.27)0.031.82 (1.25, 2.65)<0.012.91.84 (1.32, 2.56)<0.012.4Necrotizing Enterocolitis (ref = none; n = 33)0.56 (0.24, 1.29)0.170.53 (0.23, 1.24)0.142.00.68 (0.27, 1.70)0.412.3Hypoxic-Ischemic Encephalopathy (ref = none; n = 246)1.41 (0.84, 2.37)0.191.19 (0.71, 1.98)0.514.41.18 (0.87, 1.61)0.292.6Sepsis (ref = none; n = 357)1.00 (0.72, 1.39)1.001.11 (0.87, 1.43)0.392.51.14 (0.90, 1.45)0.281.8Dehydration (ref = none; n = 52)2.36 (1.14, 4.89)0.023.84 (1.34, 10.99)0.014.93.63 (1.66, 7.92)<0.012.8Single nephrotoxic medication (ref = none; n = 205)0.50 (0.32, 0.77)<0.010.46 (0.28, 0.75)<0.013.10.53 (0.39, 0.7)<0.011.9≥2 nephrotoxic medication (ref = none; n = 206)0.39 (0.24, 0.65)<0.010.39 (0.26, 0.59)<0.010.40 (0.30, 0.55)<0.01Heart Disease (ref = none; n = 32)0.55 (0.22, 1.36)0.200.49 (0.21, 1.13)0.093.30.63 (0.25, 1.58)0.323.2Vasopressor exposure (ref = none; n = 335)1.70 (1.10, 2.62)0.021.30 (0.91, 1.85)0.153.21.33 (1.05, 1.69)0.022.3Covariate reference values specified and event counts noted as “n”; Hazard ratio (HR) estimates for univariate and multivariate analyses for the complete case analysis as well as imputation model estimates are provided. In this analysis, Death and DAMA were considered competing events for AKI; Adj VIF = Average adjusted variance inflation factors. Competing risk analysis for developing AKI over time with covariates. Covariate reference values specified and event counts noted as “n”; Hazard ratio (HR) estimates for univariate and multivariate analyses for the complete case analysis as well as imputation model estimates are provided. In this analysis, Death and DAMA were considered competing events for AKI; Adj VIF = Average adjusted variance inflation factors. Death was observed in 44 neonates (7.7%) and DAMA in 160 (28.2%) neonates. AKI was significantly associated with higher death (13.5% vs. 5.3%, p < 0.01) and DAMA (34.7% vs. 25.2%, p = 0.02), and lower recovery (41.8% vs. 60.3%, p < 0.01) (Table 3). Only 9.5% of neonates were transferred to higher levels of care. Median length of hospitalisation (LOH) for the cohort was 7 days (IQR 4–13) with a shorter LOH observed in the AKI group (6 days (IQR 4–10) vs. 8 days (IQR 5–14)) compared to the non-AKI group. This difference lost statistical significance after the exclusion of patients experiencing in-hospital death or DAMA (AKI median LOH 7 days (IQR 5–14) compared to non-AKI median LOH 9 days (IQR 5–17)). A competing risk analysis found AKI not to be a significant predictor of LOH when considering death and DAMA as competing events (HR 1.00, 95% CI: 0.73–1.35) (Supplementary Table S7). There were no statistically significant differences in secondary outcomes between male and female neonates within the AKI subgroup, as well as in the overall cohort (Supplementary Table S5).Table 3Secondary outcomes among neonates with and without a diagnosis of AKI during their NICU hospitalisation.Secondary outcomesTotal (n = 570)No AKI (n = 400)AKI (n = 170)p valueDischarged against medical advice n (%)160 (28.1)101 (25.2)59 (34.7)0.02Recoverya, n (%)312 (54.7)241 (60.3)71 (41.8)<0.01Deathb, n (%)44 (7.7)21 (5.3)23 (13.5)<0.01Transfer to same/higher level of care, n (%)54 (9.5)37 (9.3)17 (10.0)0.77Length of hospitalisation (LOH) in daysEntire cohortTotal (n = 570)No AKI (n = 400)AKI (n = 170)p valueMedian LOH (IQR)7 (4–13)8 (5–14)6 (4–10)<0.01Subgroup excluding death and DAMATotal (n = 366)No AKI (n = 278)AKI (n = 88)p valueMedian LOH (IQR)9 (5–17)7 (5–14)0.08Disposition and length of hospitalisation (LOH) for the entire cohort, and LOH for a sub-group excluding neonates who experienced death and DAMA; IQR = interquartile range.aRecovery: discharged home or to lower level of care.bDeath: in-hospital mortality. Table 3Secondary outcomes among neonates with and without a diagnosis of AKI during their NICU hospitalisation.Secondary outcomesTotal (n = 570)No AKI (n = 400)AKI (n = 170)p valueDischarged against medical advice n (%)160 (28.1)101 (25.2)59 (34.7)0.02Recoverya, n (%)312 (54.7)241 (60.3)71 (41.8)<0.01Deathb, n (%)44 (7.7)21 (5.3)23 (13.5)<0.01Transfer to same/higher level of care, n (%)54 (9.5)37 (9.3)17 (10.0)0.77Length of hospitalisation (LOH) in daysEntire cohortTotal (n = 570)No AKI (n = 400)AKI (n = 170)p valueMedian LOH (IQR)7 (4–13)8 (5–14)6 (4–10)<0.01Subgroup excluding death and DAMATotal (n = 366)No AKI (n = 278)AKI (n = 88)p valueMedian LOH (IQR)9 (5–17)7 (5–14)0.08Disposition and length of hospitalisation (LOH) for the entire cohort, and LOH for a sub-group excluding neonates who experienced death and DAMA; IQR = interquartile range.aRecovery: discharged home or to lower level of care.bDeath: in-hospital mortality. Secondary outcomes among neonates with and without a diagnosis of AKI during their NICU hospitalisation. Disposition and length of hospitalisation (LOH) for the entire cohort, and LOH for a sub-group excluding neonates who experienced death and DAMA; IQR = interquartile range. Recovery: discharged home or to lower level of care. Death: in-hospital mortality. Discussion This multicentre study of outborn neonates in tele-NICUs in India demonstrates a high AKI burden with an early high-risk timeframe, after accounting for death and DAMA as competing risks. Dehydration, mechanical ventilation and vasopressor use were identified as significant predictors of early AKI after admission. Our findings were concordant with global AKI incidence rates (28–32%) thought LMICs showed wider variation.1,2,7,8 Outborn neonates are especially vulnerable to AKI.25,26 Home births, low healthcare literacy, delayed illness recognition and limited medical transport infrastructure all delay presentation to care.27,28 Clinical instabilities during transport further predispose them to hypoxia, hypoglycaemia, hypothermia and dehydration.29 Thus, our AKI incidence is likely an under-estimate of the true AKI burden in this population. Changes in serum creatinine/urine output derangements lag the onset of renal injury and are challenging to interpret in different gestations and clinical contexts.14 Frequent testing is thus essential for higher rates of AKI detection.30 In our cohort, the tele-ICU network allowed for a limited degree of uniformity in clinical recommendations at the hub through a consistent team composition. However, implementation of these recommendations at the partner sites are highly variable due to heterogeneity in baseline resources, financial capability, socio-economic status and cultural practices. These constraints led to differential testing rates across sites which is a common diagnostic challenge in India, and could have led to further under-estimation of AKI, particularly in sites with lower testing frequencies. We did not have the testing frequency data from all sites to further explore this hypothesis but acknowledge this as an important focus for future studies. Our study found a rapid time-to-diagnosis after admission, raising the possibility that renal injury preceded admission (“community-acquired AKI”). Pre-admission nephrotoxin exposure, hypoxic perinatal events and transport-related thermal or hypoxic stress can all contribute to AKI risk. However, absence of pre-admission data pertaining to birth history, prior nephrotoxin exposure, transport records and pre-admission serum creatinine values, precluded further exploration of this hypothesis. Data gaps such as these are frequent in low-resource settings in India and impede accurate risk assessment on admission. Therefore, protocolised early AKI screening and prompt institution of kidney-protective strategies at admission for outborn neonates may be considered for risk reduction. Future investigation into the extent and drivers of community-acquired AKI is also needed. Risk prediction scores and identification of high-risk periods may facilitate targeted screening and time-limited interventions among those at greatest risk, optimising the use of limited resources, while improving diagnostic rates and outcomes in resource-constrained settings.31,32 The STARZ Neonatal AKI Risk Stratification Score was developed from a multicentre prospective study of outborn and inborn neonates admitted to level 2–4 NICUs in India, and utilised limited laboratory data and clinical risk factors to predict the risk of severe AKI within 7 days of admission.33 Its high predictive performance for severe AKI and its validation in the Indian population makes it particularly promising for resource-limited settings in India.12,34 Despite several similarities between the STARZ study and ours, key differences should be considered when applying the STARZ risk stratification score to similar populations. Both the STARZ study and ours identified vasopressor use as a predictor of AKI, with directionally consistent effect estimates (our study: HR 1.33, 95% CI: 1.05–1.69; STARZ: RR 2.79, 95% CI: 1.01–2.91). Our study also identified mechanical ventilation as an independent predictor, and the STARZ study found an association between respiratory support and AKI (RR 2.62, 95% CI: 1.82–3.78).33 These findings indicate the importance of considering illness severity in neonatal AKI risk assessment. In our study, vasopressor use was independently associated with AKI in the multiply-imputed model (HR 1.33, 95% CI: 1.05–1.69, p = 0.02) but not in the complete-case analysis (HR 1.30, 95% CI: 0.91–1.85, p = 0.15). Low VIFs for vasopressor use in both models suggests that correlation with illness-severity markers do not account for this discrepancy. Given the near-identical point estimates and the narrower confidence interval achieved with imputation, this difference more likely reflects greater statistical precision from the larger effective sample size rather than bias or collinearity. Fluid balance derangements are important neonatal AKI risk factors. Prerenal AKI from dehydration is the most common cause of AKI in LMICs,12,35 particularly among outborn, transported neonates, including those with extremely low birth weight (<1500 g) or GA (<28 weeks) due to their immature skin barriers and high insensible fluid losses.36 We found that dehydration was a significant predictor of early AKI, suggesting increased vulnerability to dehydration in the community and during transport. Conversely, the STARZ study, which only enrolled neonates receiving ≥48 h of IV fluids, identified fluid overload as a significant predictor of AKI.33 Although our inclusion criteria did not have IV fluid-based limitations, and we did not assess fluid overload, yet, these complementary findings highlight the importance of monitoring and optimising fluid balance throughout the hospitalisation. Renal-pulmonary interdependence is well recognised, with AKI and fluid status closely related to pulmonary outcomes and ventilator dependence.37, 38, 39 Our finding of an association between mechanical ventilation and AKI is consistent with this. The association between optimised fluid balance and lower risks of bronchopulmonary dysplasia, NEC and PDA, shorter LOH and increased ventilator-free days,36,39, 40, 41 amplify the importance of meticulous fluid management among neonates. Nephrotoxin exposure is a recognised AKI risk factor,2,33,42 and was associated with increased AKI risk in the STARZ study (RR 2.76, 95% CI: 1.6–3.19).33 Yet, our study found that AKI was associated with lower nephrotoxin exposure. Two explanations are plausible. First, our study measured nephrotoxin exposure from admission to AKI diagnosis in the AKI group, and from admission to discharge in the non-AKI group. This asymmetry, combined with early time-to-AKI diagnosis and unrecorded pre-admission nephrotoxin exposure possibly under-estimates overall nephrotoxin exposure in the AKI group. This has been previously demonstrated.25 However, our dataset lacks the daily nephrotoxin dosage and administration data needed to confirm this with a sensitivity analysis. Second, clinicians could have avoided nephrotoxins in clinically unstable neonates perceived to be at a higher risk for AKI, resulting in confounding by indication. Unlike our study, the STARZ study identified extremely preterm infants (<28 weeks GA), sepsis, and significant cardiac disease as AKI risk factors.33 These variables along with HIE are known risk factors for AKI.1,2,43 Our population had a high representation of neonates between 30 and 36 completed weeks GA, and a notable under-representation of neonates <30 weeks GA (∼7%) and ≥37 weeks GA (∼30%). Prematurity, HIE, sepsis and congenital malformations are leading causes of neonatal mortality, particularly in resource-limited settings.44,45 It is possible that high-risk neonates such as those <28 weeks’ gestation or with HIE or cardiac disease, experience community-level mortality prior to presentation, or are referred to higher-level centres in bigger cities, and are not captured in our study. This limits the statistical power of these subgroups in our cohort and precludes the detection of a meaningful association between them and AKI. In the absence of community-level mortality and referral data from this catchment area, we could not test this hypothesis, making this explanation speculative at best. The striking male predominance (81.2%) of our cohort could reflect a referral bias stemming from a known social preference for male infants in the region that is associated with higher healthcare seeking behaviour.21 Our sex-disaggregated results found few significant baseline risk differences between these two groups, and no statistically significant differences in AKI incidence or outcomes. However, skewed male representation in our cohort may mask a higher mortality risk among female neonates due to barriers accessing care. Sex-disaggregated reporting should be encouraged in all settings, to identify healthcare inequities and inform future targeted interventions. Consistent with literature, we found that AKI was significantly associated with death1,2 in addition to DAMA. DAMA is a frequent reality in India and other similar settings, which often occurs in circumstances of critical illness with poor perceived prognosis and financial constraints,9,44,45 due to prevalent insurance and reimbursement models, health literacy, and socio-cultural practices. High DAMA rates in the AKI subgroup of our population could reflect illness severity, as evidenced by high ventilator and inotrope need in this subgroup. Given its association with post-discharge mortality,9,20,21 DAMA can serve as a proxy for overall mortality and should be considered when evaluating outcomes in India. Another important consideration of increased AKI detection through systematic screening may inadvertently influence health-seeking decisions and promote premature discharge. Intentional and nuanced counselling as well as active clinical and social support must accompany screening protocols to offset this risk. DAMA can artificially truncate hospitalisations in critically ill neonates, under-estimating AKI, LOH and in-hospital mortality measurements. In our study, the association between LOH and AKI lost statistical significance on exclusion of patients experiencing death and DAMA. Furthermore, a competing risk model showed AKI not to be a significant predictor of LOH with death and DAMA as competing risks. While this shows that AKI is likely not a driver of LOH in our cohort, we remain cautious with the interpretation of these results, as there are several factors that influence LOH in these settings including social, cultural and other unmeasured factors that cannot be disentangled from death and DAMA in our limited dataset. In regions where DAMA is a prevalent practice, it is important to record DAMA and its reasons to identify additional avenues to improve outcomes. Several important limitations must be considered. First, our strict laboratory inclusion criteria limited our eligible sample size. This together with variable laboratory testing and data entry at peripheral sites could have resulted in missed cases. Second, predictor information was captured from admission to AKI diagnosis to capture clinically relevant factors during this timeframe. However, for predictors identified late in the clinical course, this introduces temporal ambiguity as to whether a risk factor preceded AKI or co-occurred with early, subclinical renal injury. Protocolised, time-anchored screening in future studies would help clarify these temporal relationships and strengthen causal interpretation. Another limitation were the several data gaps that precluded further sub-analyses in our study. Absence of sufficient laboratory and medication details did not allow for assessment of AKI staging and recurrence, illness severity or a detailed analysis on nephrotoxin exposure on AKI. Birth and pre-hospital data representing a critical exposure window for AKI risk were also unavailable. These data are vital to the management of outborn neonates who are already at a heightened AKI risk. Additional gaps in the reasons for transfer to similar or higher level of care (from increased clinical acuity to logistical or social reasons irrespective of clinical trajectory), precluded assessment of AKI risk at transfer. Data gaps are known obstacles to gathering retrospective data in resource-constrained settings and require a systemic change with improved practices and stewardship around documentation. Lastly, despite shared oversight by a single tele-NICU team, the heterogenous implementation of clinical recommendations, inclusion of all outborn neonates, and the significant male predominance of our population, limit the generalisability of our results. There is a substantial AKI burden among outborn neonates in India, with a critical high-risk window that represents an opportunity for interventions. Our findings highlight the multifactorial nature of AKI in this population, the potential pre-admission occurrence of renal injury, and its association with death and DAMA – competing risks that are often overlooked in studies from India. Poor outcomes from neonatal AKI (increased mortality, LOH and CKD risk) are amplified and underscore the need for prompt diagnosis and management in these settings.12,31 Meticulous fluid balance management represents an immediate high-yield step to reduce AKI risk in this population. Important data gaps in pre-hospital data underscore the need for improved documentation across the clinical continuum, and future studies should focus on heterogeneity of testing frequency, as well as social determinants influencing outcomes. Policy enactments to strengthen the referral system and augmenting medical transport and infrastructure remain priorities for improving outcomes in this vulnerable population. Contributors SM: co-conceptualized the study and participated in methodology including study design, data collection and curation, data analysis, interpretation of the results, visualization (production of figures), writing – original draft and manuscript finalization. RB: participated in methodology including data collection and curation, software management and writing (review and editing) the initial and final manuscripts. ER: participated in methodology including study design, data curation and analysis, software management, visualization (production of figures), interpretation of results and writing (review and editing) of the initial and final manuscripts. CB: participated in methodology including study design, visualization (production of figures), interpretation of results and writing (review and editing) of the initial and final manuscripts. MM: was involved with study design, supervision of data curation, software management and data analysis, visualization (production of figures), interpretation of results and writing (review and editing) of the initial and final manuscripts. PYN: was involved in the methodology including study design, interpretation of results and review and editing of the initial and final manuscripts. DU: was involved in data collection and data curation, interpretation of results and review and editing of the final manuscript. GS: was involved in data collection, interpretation of results and review and editing of the final manuscript. AD: involved in study conceptualization and design, and supervised data interpretation and manuscript writing (review and editing). LG: co-conceptualized the study, and supervised funding acquisition, methodology including study design, data analysis and interpretation, and manuscript writing (review and editing) of the original draft. Data sharing statement Raw data (anonymised) is available on request. Editor note The Lancet Group takes a neutral position with respect to territorial claims in published maps and institutional affiliations. Declaration of interests SM was a paediatric resident and then neonatology fellow at Boston Children’s Hospital during the conduction of the project. LG and AD were employees at Boston Children’s Hospital and were not part of the Fred Lovejoy Resident Research and Education Award review committee that awards funding. DU is an employee at cloudphysician.ai which supported the Tele-ICU network. None of the other authors have competing interests to declare.