High Prevalence and Limited Perinatal Prediction of Kidney Dysfunction in Adolescents Born Extremely Preterm Authors: Keia R. Sanderson, Jesus E. Vazquez, Jennifer E. Flythe, David Askenazi, Michael R. Kosorok, Minxin Lu, Kartikeya Makker, Matthew Laughon, Christine E. Kistler, Rebecca C. Fry, Thomas Michael O’Shea Published: 2026 ABSTRACT IntroductionExtremely preterm birth (< 28 weeks’ gestation) increases chronic kidney disease (CKD) risk. However, among those born extremely preterm, the prevalence of abnormal kidney health during adolescence and its perinatal predictors remain poorly defined.MethodsAdolescents from the Extremely Low Gestational Age Newborn -Environmental influences on Child Health Outcomes (ELGAN-ECHO) cohort (14 US hospitals) were assessed for abnormal kidney health defined as ≥ 1 indicator—chronic kidney disease in children (CKiD) U25 estimated glomerular filtration rate (eGFR) < 90 ml/min per 1.73 m2, blood pressure (BP) ≥ 120/80 mm Hg, or microalbuminuria > 30 mg/g.Five feature selection methods screened perinatal covariate predictors using a 60 to 40 training-test split. Multiple imputation addressed missing data in sensitivity analyses. Area under the curve (AUC) was the primary performance metric.ResultsOf the 675 adolescents (mean age 18 years), 354 (52%) had a classifiable kidney health status, as follows: 188 with all 3 measurements and 166 with partial measurements showing ≥ 1 abnormality. The prevalence of abnormal kidney health was 66% among complete cases (125/188) and 82% among all determinate cases (291/354). In the full cohort, prevalence ranged from 43% to 91% (bounding estimates accounting for missing data). Individual measurements revealed 30.5% with reduced eGFR, 42% with elevated BP, and 9% with microalbuminuria. Across all methods, perinatal covariates achieved random test-set discrimination (AUC: 0.48–0.55) for abnormal kidney health prediction, compared with training set AUCs of 0.66–0.72.ConclusionAbnormal kidney health is highly prevalent after extremely preterm birth, underscoring the importance of protocolized longitudinal kidney health surveillance throughout childhood to enable early identification of CKD and elevated BP in children born extremely preterm. The perinatal factors we examined did not meaningfully predict kidney health abnormalities and variable importance did not translate to individual-level prediction in this cohort. In the US, 1 in 10 births are premature (before 37 weeks of gestation).1 Although life expectancy of individuals born preterm is only slightly reduced compared with term-born peers, preterm birth is associated with a substantially elevated lifetime risk of chronic kidney disease (CKD), with hazard ratios 2-fold higher than term born individuals.2,3 Identifying individuals born preterm at the highest risk for abnormal kidney health will support proportional intensity of surveillance, with more intensive monitoring for those at greatest risk. Such stratification could facilitate early CKD detection and interventions to slow CKD progression.4,5 Large-scale population-based studies from Europe and Japan have documented kidney disease risk across the lifespan after preterm birth; however, few studies focus specifically on kidney health during adolescence, a critical period of rapid physiologic change and potential CKD progression.6 Previous research from our group shows that 50% of adolescents born extremely preterm have at least 1 biomarker of abnormal kidney health, including microalbuminuria, reduced renal mass, and/or elevated blood pressure (BP).7 Potential mechanisms include disrupted nephrogenesis associated with preterm birth and contributions from neonatal acute kidney injury (AKI), maternal smoking, preeclampsia, maternal low socioeconomic status, and nephrotoxic neonatal antibiotic exposures.2,8, 9, 10, 11, 12, 13, 14, 15, 16 Despite the higher prevalence of CKD after preterm birth and individually recognized risk factors, whether combinations of perinatal risk factors can better predict abnormal kidney health in adolescence remains understudied.17 Consequently, no CKD prediction models currently exist for identifying at-risk individuals born preterm from neonatal intensive care unit (NICU) discharge to late adolescence. The primary objective of this study was to characterize the prevalence of abnormal kidney health in a large, multicenter cohort of adolescents born extremely preterm. The secondary objective was to determine whether perinatal risk factors, modeled with complementary machine learning approaches that accommodate high-dimensional data are sufficient to identify which individuals develop abnormal kidney health. Methods This cohort study was approved by the institutional review board of the University of North Carolina at Chapel Hill (IRB# 16-2535). ELGAN-ECHO Study Participants Study participants were a subset of the ELGAN-ECHO cohort of infants born < 28 weeks’ gestation between 2002 and 2004 at 14 US centers. Cohort details and recruitment procedures appear in previous publications.18, 19, 20 Research assistants collected perinatal covariates from medical records during NICU hospitalization. Of the 1198 survivors in the ELGAN study cohort, 675 (56%) completed at least 1 adolescent kidney health assessment (Figure 1). Adolescent assessments occurred between 2017 and 2022.Figure 1Consort Diagram of the Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes (ELGAN-ECHO) Cohort (N = 675). BP, blood pressure; NICU, neonatal intensive care unit. Consort Diagram of the Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes (ELGAN-ECHO) Cohort (N = 675). BP, blood pressure; NICU, neonatal intensive care unit. Perinatal and Adolescent Assessments A total of 473 perinatal covariates were evaluated, including maternal sociodemographic and clinical characteristics, detailed neonatal data, and placental pathology. Maternal BMI was categorized per World Health Organization definitions. Maternal covariates (age, race/ethnicity, education, Medicaid status, smoking, diabetes, and hypertension) were obtained from interview or chart review; neonatal covariates (birthweight, gestational age, biological sex assigned at birth, and clinical course) were abstracted from medical records. Kidney health was assessed once between 15 and 19 years of age and included the following 3 indicators: serum creatinine CKiD Under 25 estimated glomerular filtration rate (eGFR), BP (median of 1–3 seated oscillometric readings), and microalbuminuria (random urine albumin-to-creatinine ratio).5 Standardized Isotope Dilution Mass Spectrometry-traceable enzymatic assays measured serum creatinine, and eGFR was calculated using the CKiD U25 equation.21,22 Since the mean age at BP assessment was 18 years, the ≥ 120/80 mm Hg threshold was used to classify elevated BP according to American Academy of Pediatrics clinical practice guideline and the 2025 American College of Cardiology and American Heart Association guideline.23,24 Kidney Health Indicator Composite Outcome Definition We defined the primary outcome as the presence of ≥ 1 indicator of abnormal kidney health, as follows: serum creatinine eGFR < 90 ml/min per 1.73 m2, elevated BP ≥ 120 mm Hg (systolic) or ≥ 80 mm Hg (diastolic), and/or microalbuminuria >30 mg albumin/g creatinine.5,21, 22, 23, 24 Outcome Classification and Handling of Missing Data Of the 675 participants, 354 (52%) had a determinate kidney health status; 188 with all 3 measurements and 166 with partial measurements showed ≥ 1 abnormality. We classified the remaining 321 (48%) as indeterminate, meaning that these participants had partial kidney health measurements and all other measured indicators were normal, precluding complete outcome classification. The prevalence of abnormal kidney health was estimated by the following 3 ways: (1) determinate-case estimation using classifiable participants only, (2) bounding estimation, using the full cohort, classifying all indeterminate participants as normal (lower bound) or abnormal (upper bound), and (3) multiple imputation estimation of the full cohort in a sensitivity analysis under missing-at-random assumption. Statistical Analysis Covariate Selection We described baseline characteristics as counts and proportions for categorical covariates and as means with SD or medians with interquartile range (IQR) for continuous covariates. We retained 138 of 473 available perinatal covariates, after excluding those with substantial collinearity (r > 0.7) or more than 20% missingness, ensuring retained covariates reflected clinically observed measures and improved signal-to-noise ratio. Sample Splitting and Missing Data The dataset was randomly split into training (60%) and test (40%) sets with stratification by outcome. We addressed missing predictor values with multiple imputation by chained equations (100 imputations) performed separately within training and test sets to prevent information leakage across sets. Within each imputed dataset, the chained-equations procedure conditions variables on each other, consistent with standard practice. Feature Selection and Model Assessment We used a 2-stage analytical approach. First, 5 feature-selection methods screened 138 candidate covariates, then backward-selection models assessed final discrimination. The 5 methods were as follows: (i) full backward model selection (P< 0.05), (ii) individual backward model selection (≥ 50% selection across 100 imputed data sets), (iii) Lasso regression (10 most frequently selected via 300 bootstrap resamples, 5-fold cross-validation and 100 equally spaced λ values (0.0001–0.5), (iv) Random Forest (Gini-ranked), and (v) XGBoost (Gain-ranked).25, 26, 27 We fit ensemble methods separately to each imputed data set and aggregated importance scores across imputations.28,29 Full hyperparameter specifications appear in the Supplementary Table S1. Final Model Assessment We entered each feature-selection output into backward-selection models to determine final covariates and assess discrimination. The principal assessment used observed complete-case test-set outcomes (no imputation); imputed outcomes were evaluated in a sensitivity analysis only to test for inflation. This 2-stage design allows feature screening methods to identify potentially informative variables while using backward-selection models for the final discrimination. AUC served as the primary performance metric, categorized as random (<0.60), poor (0.60–0.69), fair (0.70–0.79), good (0.80–0.89), or excellent (≥0.90).30 Statistical Software and Reporting Descriptive statistics and tabulations were performed in SAS (version 9.4; SAS Institute, Cary, NC), and statistical analyses were performed in R (version 4.1.0; R Foundation, Vienna, Austria). Manuscript reporting followed the STROBE statement (Supplementary Table S2). Results Cohort Description In the study sample of 675 adolescents (mean age: 18 ± 0.9 years), the average maternal age at delivery was 29.6 (± 6.6) years. Maternal race and ethnicity included White (65%), Black (24%), and other (11%); 90% identified as non-Hispanic (Table 1). Nearly a quarter of birth mothers were obese (22%) and 3% had prepregnancy diabetes.Table 1Maternal and neonatal characteristics of the ELGAN-ECHO cohort (N = 675)CharacteristicOverall cohortN = 675Abnormal kidney healthaYes n = 291No n = 63Indeterminate n = 321Maternal characteristicsAge (yrs), mean (SD)29.6 (6.6)29.4 (6.7)31.8 (6.6)29.3 (6.6)Race, n (%) White438 (64.9)178 (61.2)45 (71.4)215 (67) Black158 (23.4)82 (28.2)11 (17.5)65 (20.2) Other69 (10.2)25 (8.6)7 (11.1)37 (11.5) Missing10 (1.5)6 (2.1)0 (0)4 (1.2)Hispanic, n (%)68 (10.1)26 (8.9)7 (11.1)35 (10.9) Missing3 (0.4)0 (0)0 (0)3 (0.9)BMI, n (%) < 18.552 (7.7)20 (6.9)7 (11.1)25 (7.8) ≥ 18.5 and < 30455 (67.4)184 (63.2)41 (65.1)230 (71.7) > 30144 (21.3)81 (27.8)11 (17.5)52 (16.2) Missing24 (3.6)6 (2.1)4 (6.3)14 (4.4)Chronic hypertension, n (%)101 (15)53 (18.2)7 (11.1)41 (12.8) Missing16 (2.4)3 (1)3 (4.8)10 (3.1)Preeclampsia, n (%)97 (14.4)48 (16.5)8 (12.7)41 (12.8)Smoking during pregnancy, n (%)91 (13.5)35 (12)7 (11.1)49 (15.3) Missing13 (1.9)3 (1)3 (4.8)7 (2.2)Prepregnancy diabetes, n (%)21 (3.1)13 (4.5)1 (1.6)7 (2.2) Missing15 (2.2)3 (1)3 (4.8)9 (2.8)Antenatal steroids, n (%)88 (13)38 (13.1)4 (6.3)46 (14.3)Neonatal characteristicsSex (Male), n (%)356 (52.7)152 (52.2)30 (47.6)174 (54.2)Birthweight (g), median (IQR)821 (690–960)837 (696–960)865 (705–965)800 (680–970)Gestational age (wks), n (%) 2360 (8.9)26 (8.9)7 (11.1)27 (8.4) 24127 (18.8)45 (15.5)9 (14.3)73 (22.7) 25143 (21.2)64 (22)8 (12.7)71 (22.1) 26152 (22.5)74 (25.4)18 (28.6)60 (18.7) 27193 (28.6)82 (28.2)21 (33.3)90 (28)APGAR 1 Minute, mean (SD)5 (2.2)4.8 (2.3)5.4 (2.3)5 (2.2) Missing, n (%)2 (0.3)1 (0.3)0 (0)1 (0.3)APGAR 5 Minute, mean (SD)7.1 (1.7)7 (1.7)7.4 (1.3)7 (1.7) Missing, n (%)2 (0.3)1 (0.3)0 (0)1 (0.3)SNAP score, mean (SD)21.4 (12.7)20.9 (13.1)20.9 (11.2)22.1 (12.6) Missing, n (%)11 (1.6)4 (1.4)1 (1.6)6 (1.9)Hypotensionb, n (%)150 (22.2)66 (22.7)15 (23.8)69 (21.5) Missing10 (1.5)5 (1.7)0 (0)5 (1.6)Vasopressor exposureb, n (%)254 (37.6)106 (36.4)21 (33.3)127 (39.6) Missing2 (0.3)1 (0.3)0 (0)1 (0.3)Indomethacin exposureb, n (%)411 (60.9)177 (60.8)39 (61.9)195 (60.7)Methylxanthine exposureb, n (%)491 (72.7)211 (72.5)46 (73)234 (72.9)Mechanical vent days, n (%) ≤ 7202 (29.9)99 (34)13 (20.6)90 (28) > 7 and ≤ 1480 (11.9)33 (11.3)8 (12.7)39 (12.1) > 14 and ≤ 2171 (10.5)34 (11.7)6 (9.5)31 (9.7) > 21 and ≤ 28322 (47.7)125 (43)36 (57.1)161 (50.2)Early sepsis (week 1), n (%)46 (6.8)21 (7.2)4 (6.3)21 (6.5) Missing1 (0.1)0 (0)0 (0)1 (0.3)Late sepsis (weeks 2–4), n (%)166 (24.6)61 (21)17 (27)88 (27.4) Missing5 (0.7)2 (0.7)0 (0)3 (0.9)Necrotizing enterocolitisc, n (%)51 (7.6)18 (6.2)3 (4.8)30 (9.3)Patent ductus arteriosusd, n (%)463 (68.6)195 (67)44 (69.8)224 (69.8)Bronchopulmonary dysplasia, n (%)392 (58.1)162 (55.7)39 (61.9)191 (59.5)Length of NICU stay (d), mean (SD)92.0 (39.9)91.8 (45.9)93.7 (35.7)91.8 (34.5)Discharge weight (g), median (IQR)2475 (2125–2955)2457 (2105–2905)2435 (2085–3205)2497 (2147–2957) Missing, n (%)130 (19.6)62 (21.3)9 (14.3)59 (18.4)Adolescent kidney health characteristicseGFR, median (IQR)96.6 (86.9–106.2)90.4 (81.3–102)104 (97–111.9)101.7 (95.7–114.9) Missing, n (%)321 (58.7)0 (0)0 (0)279 (86.9)Systolic blood pressure, median (IQR)116 (107–124)123 (115–130)110 (106–114)110 (103–115) Missing, n (%)176 (26.1)0 (0)0 (0)159 (49.5)Diastolic blood pressure, median (IQR)72 (66–78)76 (70–82)68 (63–72)69.5 (64–74) Missing, n (%)176 (26.1)0 (0)0 (0)159 (49.5)Microalbuminuria, median (IQR)3.2 (0–9.9)4.9 (0–17.2)2.5 (0–7.9)2.5 (0–7.2) Missing, n (%)171 (25.3)0 (0)0 (0)94 (29.3)APGAR, Appearance, Pulse, Grimace, Activity, and Respiration; ELGAN-ECHO, Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes; IQR, interquartile range.Stratified percent counts (%), means (SD), and medians (IQR) are based on the complete cases per covariates. The percentage missing on the outcome was 51.7% (349/675), whereas the percentage missing per covariate varies.aAn individual was classified as having an abnormal kidney health indicator if they met any of the following criteria: (i) estimated glomerular filtration rate (eGFR) < 90 ml/min per 1.73 m2, (ii) elevated blood pressure ≥120 mm Hg (systolic) or ≥ 80 mm Hg (diastolic), or (iii) presence of microalbuminuria > 30 mg/g. For participants where any available measured indicator was normal and ≥ 1 indicator was unmeasured are termed indeterminate.bExposure occurring within the first 14 post-natal days.cBell Stage 2b or higher necrotizing enterocolitis.dClinical or echocardiogram confirmed patent ductus arteriosus. Table 1Maternal and neonatal characteristics of the ELGAN-ECHO cohort (N = 675)CharacteristicOverall cohortN = 675Abnormal kidney healthaYes n = 291No n = 63Indeterminate n = 321Maternal characteristicsAge (yrs), mean (SD)29.6 (6.6)29.4 (6.7)31.8 (6.6)29.3 (6.6)Race, n (%) White438 (64.9)178 (61.2)45 (71.4)215 (67) Black158 (23.4)82 (28.2)11 (17.5)65 (20.2) Other69 (10.2)25 (8.6)7 (11.1)37 (11.5) Missing10 (1.5)6 (2.1)0 (0)4 (1.2)Hispanic, n (%)68 (10.1)26 (8.9)7 (11.1)35 (10.9) Missing3 (0.4)0 (0)0 (0)3 (0.9)BMI, n (%) < 18.552 (7.7)20 (6.9)7 (11.1)25 (7.8) ≥ 18.5 and < 30455 (67.4)184 (63.2)41 (65.1)230 (71.7) > 30144 (21.3)81 (27.8)11 (17.5)52 (16.2) Missing24 (3.6)6 (2.1)4 (6.3)14 (4.4)Chronic hypertension, n (%)101 (15)53 (18.2)7 (11.1)41 (12.8) Missing16 (2.4)3 (1)3 (4.8)10 (3.1)Preeclampsia, n (%)97 (14.4)48 (16.5)8 (12.7)41 (12.8)Smoking during pregnancy, n (%)91 (13.5)35 (12)7 (11.1)49 (15.3) Missing13 (1.9)3 (1)3 (4.8)7 (2.2)Prepregnancy diabetes, n (%)21 (3.1)13 (4.5)1 (1.6)7 (2.2) Missing15 (2.2)3 (1)3 (4.8)9 (2.8)Antenatal steroids, n (%)88 (13)38 (13.1)4 (6.3)46 (14.3)Neonatal characteristicsSex (Male), n (%)356 (52.7)152 (52.2)30 (47.6)174 (54.2)Birthweight (g), median (IQR)821 (690–960)837 (696–960)865 (705–965)800 (680–970)Gestational age (wks), n (%) 2360 (8.9)26 (8.9)7 (11.1)27 (8.4) 24127 (18.8)45 (15.5)9 (14.3)73 (22.7) 25143 (21.2)64 (22)8 (12.7)71 (22.1) 26152 (22.5)74 (25.4)18 (28.6)60 (18.7) 27193 (28.6)82 (28.2)21 (33.3)90 (28)APGAR 1 Minute, mean (SD)5 (2.2)4.8 (2.3)5.4 (2.3)5 (2.2) Missing, n (%)2 (0.3)1 (0.3)0 (0)1 (0.3)APGAR 5 Minute, mean (SD)7.1 (1.7)7 (1.7)7.4 (1.3)7 (1.7) Missing, n (%)2 (0.3)1 (0.3)0 (0)1 (0.3)SNAP score, mean (SD)21.4 (12.7)20.9 (13.1)20.9 (11.2)22.1 (12.6) Missing, n (%)11 (1.6)4 (1.4)1 (1.6)6 (1.9)Hypotensionb, n (%)150 (22.2)66 (22.7)15 (23.8)69 (21.5) Missing10 (1.5)5 (1.7)0 (0)5 (1.6)Vasopressor exposureb, n (%)254 (37.6)106 (36.4)21 (33.3)127 (39.6) Missing2 (0.3)1 (0.3)0 (0)1 (0.3)Indomethacin exposureb, n (%)411 (60.9)177 (60.8)39 (61.9)195 (60.7)Methylxanthine exposureb, n (%)491 (72.7)211 (72.5)46 (73)234 (72.9)Mechanical vent days, n (%) ≤ 7202 (29.9)99 (34)13 (20.6)90 (28) > 7 and ≤ 1480 (11.9)33 (11.3)8 (12.7)39 (12.1) > 14 and ≤ 2171 (10.5)34 (11.7)6 (9.5)31 (9.7) > 21 and ≤ 28322 (47.7)125 (43)36 (57.1)161 (50.2)Early sepsis (week 1), n (%)46 (6.8)21 (7.2)4 (6.3)21 (6.5) Missing1 (0.1)0 (0)0 (0)1 (0.3)Late sepsis (weeks 2–4), n (%)166 (24.6)61 (21)17 (27)88 (27.4) Missing5 (0.7)2 (0.7)0 (0)3 (0.9)Necrotizing enterocolitisc, n (%)51 (7.6)18 (6.2)3 (4.8)30 (9.3)Patent ductus arteriosusd, n (%)463 (68.6)195 (67)44 (69.8)224 (69.8)Bronchopulmonary dysplasia, n (%)392 (58.1)162 (55.7)39 (61.9)191 (59.5)Length of NICU stay (d), mean (SD)92.0 (39.9)91.8 (45.9)93.7 (35.7)91.8 (34.5)Discharge weight (g), median (IQR)2475 (2125–2955)2457 (2105–2905)2435 (2085–3205)2497 (2147–2957) Missing, n (%)130 (19.6)62 (21.3)9 (14.3)59 (18.4)Adolescent kidney health characteristicseGFR, median (IQR)96.6 (86.9–106.2)90.4 (81.3–102)104 (97–111.9)101.7 (95.7–114.9) Missing, n (%)321 (58.7)0 (0)0 (0)279 (86.9)Systolic blood pressure, median (IQR)116 (107–124)123 (115–130)110 (106–114)110 (103–115) Missing, n (%)176 (26.1)0 (0)0 (0)159 (49.5)Diastolic blood pressure, median (IQR)72 (66–78)76 (70–82)68 (63–72)69.5 (64–74) Missing, n (%)176 (26.1)0 (0)0 (0)159 (49.5)Microalbuminuria, median (IQR)3.2 (0–9.9)4.9 (0–17.2)2.5 (0–7.9)2.5 (0–7.2) Missing, n (%)171 (25.3)0 (0)0 (0)94 (29.3)APGAR, Appearance, Pulse, Grimace, Activity, and Respiration; ELGAN-ECHO, Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes; IQR, interquartile range.Stratified percent counts (%), means (SD), and medians (IQR) are based on the complete cases per covariates. The percentage missing on the outcome was 51.7% (349/675), whereas the percentage missing per covariate varies.aAn individual was classified as having an abnormal kidney health indicator if they met any of the following criteria: (i) estimated glomerular filtration rate (eGFR) < 90 ml/min per 1.73 m2, (ii) elevated blood pressure ≥120 mm Hg (systolic) or ≥ 80 mm Hg (diastolic), or (iii) presence of microalbuminuria > 30 mg/g. For participants where any available measured indicator was normal and ≥ 1 indicator was unmeasured are termed indeterminate.bExposure occurring within the first 14 post-natal days.cBell Stage 2b or higher necrotizing enterocolitis.dClinical or echocardiogram confirmed patent ductus arteriosus. Maternal and neonatal characteristics of the ELGAN-ECHO cohort (N = 675) APGAR, Appearance, Pulse, Grimace, Activity, and Respiration; ELGAN-ECHO, Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes; IQR, interquartile range. Stratified percent counts (%), means (SD), and medians (IQR) are based on the complete cases per covariates. The percentage missing on the outcome was 51.7% (349/675), whereas the percentage missing per covariate varies. An individual was classified as having an abnormal kidney health indicator if they met any of the following criteria: (i) estimated glomerular filtration rate (eGFR) < 90 ml/min per 1.73 m2, (ii) elevated blood pressure ≥120 mm Hg (systolic) or ≥ 80 mm Hg (diastolic), or (iii) presence of microalbuminuria > 30 mg/g. For participants where any available measured indicator was normal and ≥ 1 indicator was unmeasured are termed indeterminate. Exposure occurring within the first 14 post-natal days. Bell Stage 2b or higher necrotizing enterocolitis. Clinical or echocardiogram confirmed patent ductus arteriosus. Among the neonatal characteristics, 53% were of male sex, the median birthweight was 821 (IQR 690–960) grams, and the median gestational age was 26 (IQR 24–27) weeks. Thirty-eight percent of participants received vasopressors, 61% received indomethacin, 73% received a methylxanthine in the first 14 postnatal days, 48% required mechanical ventilator support for ≥ 21 cumulative postnatal days, and 49% were diagnosed with bronchopulmonary dysplasia. The mean duration of neonatal hospitalization was 92 (± 40) days. Prevalence Estimates of Adolescent Kidney Health After Extremely Preterm Birth Of the 675 adolescents, 354 (52%) had a classifiable kidney health status: 188 had all 3 measurements (125 abnormal, 63 normal) and 166 had partial measurements showing ≥ 1 abnormality (Table 2). Among the 188 with complete assessments, 66% (125/188) had ≥ 1 abnormal indicator. When we included all 354 determinate cases, (188 complete + 166 with partial measurements showing ≥ 1 abnormality), 82% (291/354) had ≥ 1 abnormal indicator. The higher prevalence in the full determinate sample reflects the inclusion of the 166 participants with partial measurements showing ≥ 1 abnormality, who are classified as abnormal by definition (Supplementary Table S3).Table 2Adolescent kidney health outcomes of the ELGAN-ECHO cohort (N = 675)OutcomeFull study sample frequency (%)Abnormal kidney healtha Yes291 (43.1) No63 (9.3) Missing data for one or more indicators of kidney health321 (47.6)Abnormal estimated glomerular filtration rate (eGFR) Yes85 (12.6) No194 (28.7) Missing data for eGFR396 (58.7)Elevated blood pressure (BP) Yes210 (31.1) No289 (42.8) Missing data for BP176 (26.1)Microalbuminuria Yes46 (6.8) No458 (67.9) Missing data for microalbuminuria171 (25.3)Age at kidney health follow up (years), mean (SD)18.3 (0.9)ELGAN-ECHO, Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes.aAbnormal kidney health was defined having one or more of the following : (i) estimated glomerular filtration rate (eGFR) < 90 ml/min per 1.73 m2, (ii) elevated blood pressure ≥ 120 mm Hg (systolic) or ≥ 80 mm Hg (diastolic), or (iii) presence of microalbuminuria > 30 mg/g. Table 2Adolescent kidney health outcomes of the ELGAN-ECHO cohort (N = 675)OutcomeFull study sample frequency (%)Abnormal kidney healtha Yes291 (43.1) No63 (9.3) Missing data for one or more indicators of kidney health321 (47.6)Abnormal estimated glomerular filtration rate (eGFR) Yes85 (12.6) No194 (28.7) Missing data for eGFR396 (58.7)Elevated blood pressure (BP) Yes210 (31.1) No289 (42.8) Missing data for BP176 (26.1)Microalbuminuria Yes46 (6.8) No458 (67.9) Missing data for microalbuminuria171 (25.3)Age at kidney health follow up (years), mean (SD)18.3 (0.9)ELGAN-ECHO, Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes.aAbnormal kidney health was defined having one or more of the following : (i) estimated glomerular filtration rate (eGFR) < 90 ml/min per 1.73 m2, (ii) elevated blood pressure ≥ 120 mm Hg (systolic) or ≥ 80 mm Hg (diastolic), or (iii) presence of microalbuminuria > 30 mg/g. Adolescent kidney health outcomes of the ELGAN-ECHO cohort (N = 675) ELGAN-ECHO, Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes. Abnormal kidney health was defined having one or more of the following : (i) estimated glomerular filtration rate (eGFR) < 90 ml/min per 1.73 m2, (ii) elevated blood pressure ≥ 120 mm Hg (systolic) or ≥ 80 mm Hg (diastolic), or (iii) presence of microalbuminuria > 30 mg/g. In the full cohort of 675 participants, the bounding estimate placed abnormal kidney health prevalence between 43% (lower bound, assuming all 321 indeterminate cases were normal) and 91% (upper bound, assuming all were abnormal). Under a missing-at-random assumption, multiple imputation estimated prevalence at 70% (95% CI: 65.3%–74.3%; Supplementary Table S4). Among the participants with individual measurements, 30.5% had reduced eGFR (85/279), 42% had elevated BP (210/499), and 9% had microalbuminuria (46/504). Among all adolescents with ≥ 1 measured indicator, the median values were as follows: eGFR was 90 (IQR: 81–102) ml/min per 1.73 m2; systolic BP was 123 (IQR: 115– 130) mm Hg, diastolic BP was 76 (IQR: 70–82) mm Hg, and urine microalbumin was 4.9 (IQR: 0–17) mg/g. Feature Selection Analysis The data set was split into training (60%, n = 405) and test (40%, n = 270) sets (Table 3). Test-set discrimination on observed complete-case outcomes (primary assessment) showed that perinatal covariates achieved random discrimination for abnormal kidney health (AUC: 0.48–0.55), indicating no clinically meaningful predictive performance. Training-set AUCs were higher (0.66–0.72), demonstrating overfitting (Table 4). Complete-case test-set results and multiple-imputation sensitivity analyses were concordant, with no meaningful improvement from imputation.Table 3Distribution of abnormal kidney health indicator after sample splitting, ELGAN-ECHO cohort (N = 675)Abnormal kidney health indicatoraTraining (n = 405)Test (n = 270)Yes (n = 291)171 (42.2)120 (44.4)No (n = 63)42 (10.4)21 (7.8)Missing (n = 321)192 (47.4)129 (47.8)ELGAN-ECHO, Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes.Sample split ratio 60:40.aAn individual was classified as having an abnormal kidney health indicator if they met any of the following criteria: (i) estimated glomerular filtration rate (eGFR) < 90 ml/min per 1.73 m2, (ii) elevated blood pressure ≥ 120 mm Hg (systolic) or ≥ 80 mm Hg (diastolic), or (iii) presence of microalbuminuria > 30 mg/g.Table 4Machine learning model performance as measured by the AUC from the ELGAN-ECHO cohortOutcomeaModelAUCbTraining setTest set (multiple imputation)Test setc (no imputation)Abnormal kidney healthBackward MS0.660.530.55Individual Backward MS0.720.530.50Lasso0.680.520.48Random Forest0.700.520.55XGBoost0.680.520.48Abnormal eGFRBackward MS0.620.530.55Individual Backward MS0.620.530.55Lasso0.620.530.55Random Forest0.610.530.55XGBoost0.610.530.55Elevated blood pressureBackward MS0.740.530.55Individual Backward MS0.770.540.58Lasso0.650.510.55Random Forest0.720.530.61XGBoost0.710.520.53MicroalbuminuriaBackward MS0.720.570.60Individual Backward MS0.720.560.60Lasso0.700.530.68Random Forest0.720.550.73XGBoost0.680.550.67AUC, area under the curve; ELGAN-ECHO, Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes.Sample split ratio 60:40.aAn individual was classified as having an abnormal kidney health indicator if they met any of the following criteria: (i) estimated glomerular filtration rate (eGFR) < 90 ml/min per 1.73 m2, (ii) elevated blood pressure ≥ 120 mm Hg (systolic) or ≥ 80 mm Hg (diastolic), or (iii) presence of microalbuminuria > 30 mg/g.bAUCs were calculated separately for each imputed dataset (100 total), and the mean AUC was calculated.cWhen no imputation was performed, the single AUC was reported. Test set sample size (without imputation) varies by outcome—abnormal kidney health indicator (n = 114, outcome count is 97/114), abnormal eGFR (n = 92, outcome count is 29), microalbuminuria (n = 161, outcome count is 8), elevated blood pressure (n = 122, outcome count is 57). For microalbuminuria, the low event count (n = 8) renders the AUC estimate imprecise. Table 3Distribution of abnormal kidney health indicator after sample splitting, ELGAN-ECHO cohort (N = 675)Abnormal kidney health indicatoraTraining (n = 405)Test (n = 270)Yes (n = 291)171 (42.2)120 (44.4)No (n = 63)42 (10.4)21 (7.8)Missing (n = 321)192 (47.4)129 (47.8)ELGAN-ECHO, Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes.Sample split ratio 60:40.aAn individual was classified as having an abnormal kidney health indicator if they met any of the following criteria: (i) estimated glomerular filtration rate (eGFR) < 90 ml/min per 1.73 m2, (ii) elevated blood pressure ≥ 120 mm Hg (systolic) or ≥ 80 mm Hg (diastolic), or (iii) presence of microalbuminuria > 30 mg/g. Distribution of abnormal kidney health indicator after sample splitting, ELGAN-ECHO cohort (N = 675) ELGAN-ECHO, Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes. Sample split ratio 60:40. An individual was classified as having an abnormal kidney health indicator if they met any of the following criteria: (i) estimated glomerular filtration rate (eGFR) < 90 ml/min per 1.73 m2, (ii) elevated blood pressure ≥ 120 mm Hg (systolic) or ≥ 80 mm Hg (diastolic), or (iii) presence of microalbuminuria > 30 mg/g. Table 4Machine learning model performance as measured by the AUC from the ELGAN-ECHO cohortOutcomeaModelAUCbTraining setTest set (multiple imputation)Test setc (no imputation)Abnormal kidney healthBackward MS0.660.530.55Individual Backward MS0.720.530.50Lasso0.680.520.48Random Forest0.700.520.55XGBoost0.680.520.48Abnormal eGFRBackward MS0.620.530.55Individual Backward MS0.620.530.55Lasso0.620.530.55Random Forest0.610.530.55XGBoost0.610.530.55Elevated blood pressureBackward MS0.740.530.55Individual Backward MS0.770.540.58Lasso0.650.510.55Random Forest0.720.530.61XGBoost0.710.520.53MicroalbuminuriaBackward MS0.720.570.60Individual Backward MS0.720.560.60Lasso0.700.530.68Random Forest0.720.550.73XGBoost0.680.550.67AUC, area under the curve; ELGAN-ECHO, Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes.Sample split ratio 60:40.aAn individual was classified as having an abnormal kidney health indicator if they met any of the following criteria: (i) estimated glomerular filtration rate (eGFR) < 90 ml/min per 1.73 m2, (ii) elevated blood pressure ≥ 120 mm Hg (systolic) or ≥ 80 mm Hg (diastolic), or (iii) presence of microalbuminuria > 30 mg/g.bAUCs were calculated separately for each imputed dataset (100 total), and the mean AUC was calculated.cWhen no imputation was performed, the single AUC was reported. Test set sample size (without imputation) varies by outcome—abnormal kidney health indicator (n = 114, outcome count is 97/114), abnormal eGFR (n = 92, outcome count is 29), microalbuminuria (n = 161, outcome count is 8), elevated blood pressure (n = 122, outcome count is 57). For microalbuminuria, the low event count (n = 8) renders the AUC estimate imprecise. Machine learning model performance as measured by the AUC from the ELGAN-ECHO cohort AUC, area under the curve; ELGAN-ECHO, Extremely Low Gestational Age Newborn-Environmental Child Health Outcomes. Sample split ratio 60:40. An individual was classified as having an abnormal kidney health indicator if they met any of the following criteria: (i) estimated glomerular filtration rate (eGFR) < 90 ml/min per 1.73 m2, (ii) elevated blood pressure ≥ 120 mm Hg (systolic) or ≥ 80 mm Hg (diastolic), or (iii) presence of microalbuminuria > 30 mg/g. AUCs were calculated separately for each imputed dataset (100 total), and the mean AUC was calculated. When no imputation was performed, the single AUC was reported. Test set sample size (without imputation) varies by outcome—abnormal kidney health indicator (n = 114, outcome count is 97/114), abnormal eGFR (n = 92, outcome count is 29), microalbuminuria (n = 161, outcome count is 8), elevated blood pressure (n = 122, outcome count is 57). For microalbuminuria, the low event count (n = 8) renders the AUC estimate imprecise. In secondary analyses of individual indicators of kidney health, training set AUCs exceeded 0.70 for elevated BP and microalbuminuria under several methods; however, test-set discrimination remained random or poor for all indicators. For elevated BP, only the Random Forest model achieved an AUC > 0.60 in the test set (observed outcomes). For microalbuminuria, only the Random Forest model exceeded an AUC above 0.70 in the test set; however, the low event rate (8/161) renders the estimate imprecise. Across all 5 feature selection methods (full backward elimination, individual backward elimination, Lasso, Random Forest, XGBoost), the same core perinatal variables were consistently identified as important—maternal prepregnancy obesity, multiple gestation, high partial pressure of carbon dioxide (pCO2) on postnatal day 1, total mechanical ventilation days within the first 28 postnatal days, and gestational age at birth (Figure 2).Figure 2(a–d) Selection frequency of the top 10 covariates across 100 imputed data sets, by outcome and method. eGFR, estimated glomerular filtration rate; NICU, neonatal intensive care unit; PND, postnatal day; PROM, premature rupture of membranes. (a–d) Selection frequency of the top 10 covariates across 100 imputed data sets, by outcome and method. eGFR, estimated glomerular filtration rate; NICU, neonatal intensive care unit; PND, postnatal day; PROM, premature rupture of membranes. Discussion In this large, multicenter cohort of US adolescents born extremely preterm, abnormal kidney health was common, with a prevalence of 66% among those with complete assessments (125/188 participants) and 82% among all classifiable cases (291/354 participants). In the full cohort, prevalence estimates ranged from 43% (lower bound, assuming all 321 indeterminate cases were normal) and 91% (upper bound, assuming all were abnormal), with 70% under missing-at-random assumptions. Perinatal factors, examined with 5 feature-selection methods to interrogate 138 perinatal covariates, did not identify which individuals had abnormal kidney health. Test-set discrimination for the composite outcome was random, providing no clinically meaningful individual-level predictive performance. These findings demonstrate that perinatal factors captured in this dataset were insufficient to achieve clinical discrimination for individual kidney risk stratification in later childhood. This does not preclude the existence of other meaningful perinatal predictors; rather, it suggests that identifying at-risk infants at NICU discharge may require additional variables not captured in this cohort, alongside longitudinal approaches that incorporate childhood exposures. The prevalence of abnormal kidney health in the present sample (66% among complete cases) substantially exceeds estimates from population-based studies of preterm birth, though few large-scale cohort studies report the prevalence of kidney abnormalities during adolescence in extremely preterm survivors. Crump and colleagues reported clinical CKD in only 0.19% of extremely preterm children ages 10 to 19 (incidence 24.13 per 100,000 person-years).2 Gjerde et al.31,32 found that small for gestational age and low birth weight carried a cumulative incidence of end-stage kidney disease of 0.014%, by age 20 years. Hirano et al.33 reported a CKD prevalence of 0.006% (5.97 per 100,000) among children ages 3 months to 15 years who were born low birth weight. The higher prevalence in our study reflects several methodological distinctions, as follows: (i) our use of a broader outcome definition, encompassing elevated BP and microalbuminuria in addition to reduced eGFR, (ii) restriction of our sample to extremely preterm births (< 28 weeks), the highest-risk preterm subset; and (iii) our use of standardized laboratory assessments and oscillometric BP measurements to identify subclinical kidney dysfunction, rather than relying solely on clinical diagnoses. Both perspectives are clinically relevant. Registry data establish that preterm birth increases lifetime CKD risk, whereas our findings emphasize the high frequency of subclinical kidney abnormalities as early as adolescence among individuals born extremely preterm. Notably, only 52% of this high-risk cohort had a classifiable kidney health status (354 of 675); 48% had indeterminate status because of missing measurements. The 354 determinate cases may differ systematically from the 321 indeterminate participants, introducing potential selection bias. The bounding estimate (43%–91%) is valid under any missingness mechanism, including missing not at random, and does not assume homogeneity between groups. To address this potential bias, we used the following 3 complementary prevalence approaches with different assumptions: complete-case (66%), bounding estimates (43%–91%), and imputation (70%). This range allows evaluation of the prevalence across plausible assumptions regarding missing data. Regardless of the inability of perinatal factors to predict individual risk, the high prevalence of abnormal kidney health in adolescents born extremely preterm serves as a call to action for protocolized longitudinal surveillance extending well beyond current consensus recommendations (at the age of 2 years) to enable early CKD detection and intervention.34 It also highlights an opportunity to raise community awareness of the importance of kidney health testing in high-risk children in the same way that efforts have been made to encourage kidney health testing in adults. This study represents the first application of 5 analytically distinct methods to screen 138 perinatal covariates for CKD prediction among children born extremely prematurely. The same 5 perinatal factors were consistently identified—maternal prepregnancy obesity, multiple gestation, lower gestational age at birth, prolonged mechanical ventilation, and neonatal hypercarbia. However, discrimination collapsed from training (AUC: 0.66–0.72) to test-set performance (AUC: 0.52–0.53), indicating that, although these perinatal variables may be indepdently associated with kidney health abnormalities, they were insufficient for predicting individual kidney health status in the validation sample. Complete-case test-set results (observed outcomes, no imputation) were the principal assessment; multiple-imputation results were comparable (AUC: 0.52–0.53), confirming that imputation did not artificially inflate discrimination. In secondary analyses of individual kidney health indicators, training-set AUCs exceeded 0.70 for elevated BP and microalbuminuria; however, test-set discrimination remained poor for all indicators, with only Random Forest achieving AUC > 0.60 for elevated BP. Several limitations should be acknowledged. First, cross-sectional assessment of kidney health at a single time point prevents evaluation of disease trajectory or determination of when abnormalities emerge. In addition, glomerular hyperfiltration may have masked early kidney disease in some adolescents.35,36 Second, single-visit BP measurement using oscillometric readings may not capture usual BP, introducing misclassification from white-coat or masked hypertension. Third, random microalbuminuria samples, rather than first morning urine samples may overestimate the prevalence because of orthostatic proteinuria, particularly in adolescents. Fourth, 48% of participants had indeterminate kidney health data. This introduces potential selection bias, as those with complete data may differ systematically from those with missing indicators. Fifth, the cohort is restricted to extremely preterm births (< 28 weeks), limiting generalizability to the broader preterm population. Sixth, we evaluated only exposures during NICU hospitalization (averaging 13 postnatal weeks), yet the gap between discharge and adolescent assessment spans 15 to 19 years during which unmeasured “second hits” likely occur. Post-NICU exposures including childhood obesity, nephrotoxic medications, dietary sodium intake, and social determinants of health may exert effects equal to or greater than perinatal factors, introducing variance that birth hospitalization data alone cannot capture. Seventh, congenital kidney anomalies and acute kidney injury were not rigorously assessed, so we cannot be certain that the study sample was free of such abnormalities. Conclusion The high prevalence of abnormal kidney health in this cohort of extremely preterm adolescents underscores the need for protocolized longitudinal kidney health surveillance throughout childhood to enable early identification of CKD. The perinatal factors we analyzed did not identify which adolescents were affected, suggesting that kidney health risk stratification at NICU discharge will require additional data sources. Future research should identify additional data sources and childhood factors that improve kidney health risk prediction after preterm birth and develop protocolized surveillance pathways to enable early identification and intervention in children born extremely preterm. Disclosure In the last 24 months, DJA has consulted with Nuwellis and Seastar. He has financial interests in patents and/or innovations on a noninvasive urine collection and system to monitor, prescribe, and deliver cumulative fluid balance. He is the Founder and Chief Scientific Officer for Zorro-Flow Inc. The other authors have no conflicts of interest relevant to this article to disclose.