Uncovering Hidden Phenotypes of Bronchopulmonary Dysplasia: A Latent Class Analysis of Preterm Infants in a Middle-Income Setting

Uncovering Hidden Phenotypes of Bronchopulmonary Dysplasia: A Latent Class Analysis of Preterm Infants in a Middle-Income Setting | Authorea try { document.documentElement.classList.add('js'); } catch (e) { } var _gaq = _gaq || []; _gaq.push(['_setAccount', 'G-8VDV14Y67G']); _gaq.push(['_trackPageview']); (function() { var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true; ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js'; var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s); })(); Skip to main content Preprints Collections Wiley Open Research IET Open Research Ecological Society of Japan All Collections About About Authorea FAQs Contact Us Quick Search anywhere Search for preprint articles, keywords, etc. Search Search ADVANCED SEARCH SCROLL This is a preprint and has not been peer reviewed. Data may be preliminary. 11 November 2025 V1 Latest version Share on Uncovering Hidden Phenotypes of Bronchopulmonary Dysplasia: A Latent Class Analysis of Preterm Infants in a Middle-Income Setting Authors : Andrea Parra Buitrago 0009-0002-9665-5791 [email protected] , Andrea Jaramillo Cerezo , and Jefferson Buendia 0000-0003-2404-6612 Authors Info & Affiliations https://doi.org/10.22541/au.176289131.11095672/v1 236 views 172 downloads Contents Abstract Information & Authors Metrics & Citations View Options References Figures Tables Media Share Abstract Background Bronchopulmonary dysplasia (BPD) is a heterogeneous syndrome encompassing multiple biological and structural phenotypes. This study aimed to identify latent phenotypic subgroups of preterm infants with BPD and characterize their clinical and perinatal profiles. Methods We conducted a retrospective cohort of preterm infants born at ≤32 weeks’ gestation and admitted to a tertiary neonatal unit in Medellín, Colombia (2021–2023). Infants with major malformations or genetic syndromes were excluded. Clinical and respiratory variables were analyzed using latent class analysis (LCA) to identify subgroups based on markers of prematurity, oxygen dependency, and respiratory support. Results Among 236 infants, a three-class model best described the data (AIC = 12,002; BIC = 12,440; entropy = 0.83). Class 1 (64%) comprised more mature infants with mild respiratory disease and 5% mortality. Class 2 (27%) showed prolonged oxygen therapy and intermediate outcomes. Class 3 (9%), composed mainly of extremely preterm infants, exhibited severe respiratory failure and 88% mortality. The classes reflected a progression from mild inflammatory to interstitial and vascular phenotypes. Conclusions LCA revealed three distinct BPD phenotypes with increasing respiratory severity and mortality, paralleling biological models of parenchymal, interstitial, and vascular injury. This approach supports data-driven phenotyping as a foundation for precision neonatal care. Uncovering Hidden Phenotypes of Bronchopulmonary Dysplasia: A Latent Class Analysis of Preterm Infants in a Middle-Income Setting Short running title: Phenotypes of Bronchopulmonary Dysplasia Andrea Parra Buitrago MD 1 , Andrea Jaramillo Cerezo MD 1 , Jefferson Antonio Buendía, MD PhD 2 Universidad Pontificia Bolivariana. Medellín, Colombia Research group in Pharmacology and Toxicology. Department of Pharmacology and Toxicology. University of Antioquia, Medellín, Colombia Corresponding author: Andrea Parra Buitrago, Universidad Pontificia Bolivariana. Medellín, Colombia. [email protected] Background Bronchopulmonary dysplasia (BPD) is a heterogeneous syndrome encompassing multiple biological and structural phenotypes. This study aimed to identify latent phenotypic subgroups of preterm infants with BPD and characterize their clinical and perinatal profiles. Methods We conducted a retrospective cohort of preterm infants born at ≤32 weeks’ gestation and admitted to a tertiary neonatal unit in Medellín, Colombia (2021–2023). Infants with major malformations or genetic syndromes were excluded. Clinical and respiratory variables were analyzed using latent class analysis (LCA) to identify subgroups based on markers of prematurity, oxygen dependency, and respiratory support. Results Among 236 infants, a three-class model best described the data (AIC = 12,002; BIC = 12,440; entropy = 0.83). Class 1 (64%) comprised more mature infants with mild respiratory disease and 5% mortality. Class 2 (27%) showed prolonged oxygen therapy and intermediate outcomes. Class 3 (9%), composed mainly of extremely preterm infants, exhibited severe respiratory failure and 88% mortality. The classes reflected a progression from mild inflammatory to interstitial and vascular phenotypes. Conclusions LCA revealed three distinct BPD phenotypes with increasing respiratory severity and mortality, paralleling biological models of parenchymal, interstitial, and vascular injury. This approach supports data-driven phenotyping as a foundation for precision neonatal care. Introduction Bronchopulmonary dysplasia (BPD) remains the most common serious respiratory morbidity in preterm infants, affecting up to 45% of those born before 28 weeks of gestation (1). Despite advances in neonatal care—including less invasive ventilation strategies and use of postnatal steroids—the incidence of BPD has remained stable or increased due to improved survival of extremely preterm infants (2). However, BPD is increasingly recognized not as a single entity but as a heterogeneous disorder resulting from multiple converging pathogenic pathways including lung immaturity, ventilator-induced lung injury, infection/inflammation, and disrupted vascular development (3). This complexity necessitates a deeper understanding of the clinical and pathobiological phenotypes that underlie BPD and its long-term sequelae. In recent years, latent class analysis (LCA) has emerged as a robust, person-centered, data-driven method to identify unobserved subgroups (latent classes) of patients based on observed indicators such as neonatal comorbidities, respiratory support, and growth restriction (4). Several studies have demonstrated the utility of LCA to understanding of neonatal risk factors in neurodevelopment (5) or to identify patterns of health, socioeconomic, behavioral, and psychosocial indicators that may be associated with low birth weight delivery or preterm birth(6). This approach allows for more precise patient stratification, enabling the development of targeted interventions and prognostic tools that go beyond traditional severity grading systems. Nonetheless, most LCA studies on preterms have been conducted in high-income countries (HICs), where access to advanced neonatal care is widespread and environmental exposures are relatively controlled(7). In contrast, low- and middle-income countries (LMICs), especially those in tropical regions, face a higher burden of preterm birth (10–12% vs. exposure to air pollution (including biomass fuel), higher infection rates, and inconsistent access to evidence-based interventions (8, 9). For example, in LMIC settings, neonatal sepsis—a known risk factor for BPD—is significantly more prevalent and associated with a 4.15-fold increased risk of BPD(10) . In addition, long-term outcomes may be influenced by environmental factors such as tropical viral epidemics, malnutrition, and altered patterns of post-discharge care. Given these differences, it is plausible that BPD phenotypes in LMICs differ not only in prevalence but also in underlying mechanisms and prognostic implications. Understanding these distinct phenotypes in tropical and resource-limited contexts is essential to develop locally adapted prediction models, surveillance tools, and therapeutic strategies. In this study, we aimed to identify latent phenotypes of BPD using LCA in a neonatal cohort from a tertiary center in a tropical LMIC setting. We hypothesized that distinct clinical subgroups would emerge based on perinatal risk factors and early respiratory morbidity, and that these subgroups would differ in key neonatal outcomes. This work contributes to the growing global literature on BPD phenotyping and addresses the gap in knowledge from underrepresented geographic and socioeconomic settings. Materials and Methods Study design and setting This study was designed as a retrospective cohort conducted at the neonatal intensive care unit (NICU) of Clínica Universitaria Bolivariana, a tertiary referral hospital located in Medellín, Colombia. The study was conceived to explore phenotypic patterns and clinical outcomes among preterm infants with varying degrees of respiratory morbidity, including bronchopulmonary dysplasia (BPD). Ethical Approval and Informed Consent This study was approved by the Ethics Committee for Health Research of the Universidad Pontificia Bolivariana (Medellín, Colombia) under Act No. 19 of 2025, reference CEI-0283-OCT-2025. The committee reviewed and approved the study protocol titled “Phenotypes associated with the development of adverse pulmonary outcomes in patients with bronchopulmonary dysplasia at a reference center in Medellín, Colombia, between 2021 and 2023.” Given the retrospective design, the requirement for informed consent was waived by the institutional ethics committee, as all data were obtained from de-identified medical records and no direct contact with patients or families occurred. All procedures complied with the Declaration of Helsinki, the Colombian Ministry of Health Resolution 8430 of 1993, and relevant institutional ethical standards. Patient selection and eligibility criteria The recruitment process was conducted in several stages. Initially, the research team obtained from the hospital information system a complete registry of all neonates admitted to the NICU between 2021 and 2022. Because the electronic records did not allow automatic filtering by gestational age, the investigators manually reviewed the entire list and selected exclusively infants born at ≤ 32 weeks of gestation, who constituted the primary inclusion population. Infants were excluded if they met any of the following conditions: (1) presence of major congenital malformations or confirmed genetic syndromes; (2) care under palliative or comfort-only management plans; (3) transfer to another hospital during the index hospitalization with no subsequent readmission to the Bolivariana NICU, which precluded longitudinal follow-up; or (4) birth at an external institution, given the limited accuracy of perinatal and maternal data in such cases. The cohort was subsequently expanded to include additional preterm infants admitted during 2023, following updated approval from the ethics committee to increase statistical power and representation of recent clinical practices. All eligible patients meeting the inclusion criteria and with sufficient perinatal and respiratory data were retained for analysis. After exclusions and data quality verification, the final analytical dataset comprised 236 neonates. Variables and data collection Clinical information was extracted from the electronic medical records by trained investigators using a standardized data collection form. Variables were selected according to their clinical relevance and prior evidence linking them to the development and severity of bronchopulmonary dysplasia. These variables were categorized into four domains: (1) Perinatal and maternal factors, including maternal age, antenatal care, rupture of membranes, gestational age by early obstetric ultrasound, and small-for-gestational-age status; (2) Respiratory support and oxygenation, including type and duration of mechanical ventilation (invasive or non-invasive), days on oxygen therapy, and need for supplemental oxygen at 36 weeks postmenstrual age; (3) Infectious exposures, including clinical or laboratory-confirmed neonatal infection and suspected chorioamnionitis; and (4) Comorbidities, such as persistence of ductus arteriosus at 36 weeks, pulmonary hypertension, retinopathy of prematurity, and other systemic complications. Study design rationale The retrospective cohort design allowed reconstruction of the clinical trajectories of preterm infants during their NICU hospitalization, facilitating a multidimensional characterization of respiratory evolution. This design was particularly appropriate for identifying latent subgroups (phenotypes) of BPD through secondary analysis of routinely collected data, while minimizing selection bias and ensuring representativeness of local neonatal populations in middle-income settings. Data Analytic Plan Continuous variables were assessed for normality and distributional consistency. To enhance interpretability and reduce skewness, several variables (e.g., duration of oxygen therapy and gestational age) were standardized or transformed into quintiles when appropriate. Binary variables were defined based on clinically established cutoffs. Variables with partial missingness were imputed using multiple imputation by chained equations (MICE), applying predictive mean matching for continuous variables and logistic or ordinal logistic models for binary or ordinal variables. Ten imputed datasets were generated and combined following Rubin’s rules to ensure robust estimation. The main objective of the analysis was to identify subgroups (latent classes) of preterm infants with distinct patterns of early life exposures and respiratory morbidity using latent class analysis (LCA). Models with one to five latent classes were fitted using generalized structural equation modeling (GSEM) in Stata 18.0. Variables were specified with their appropriate distribution families and link functions: binomial with logit link for binary outcomes, ordinal logistic for ordinal variables (in models tested), and Gaussian with log link for continuous variables. All models assumed local independence among indicators within each class. Latent classes were defined as categorical latent variables with fixed intercepts only (no covariates), and no constraints were imposed on item-response probabilities across classes. Model fit was evaluated using multiple criteria: log-likelihood, Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and entropy. Entropy values ≥0.8 were considered indicative of good class separation. Additionally, the average posterior probability of class membership for each individual was inspected to ensure accurate classification. Solutions with fewer than 5% of the total sample in any class were considered unstable. The number of classes was selected based on a combination of statistical indicators, parsimony, interpretability, and clinical relevance. Once the optimal model was identified, class membership was assigned using the maximum posterior probability for each patient. Classes were then compared in terms of baseline variables and clinical outcomes (including mortality, BPD severity, length of stay, and use of postnatal steroids) using Chi-square or Fisher’s exact tests for categorical variables and ANOVA or Kruskal–Wallis tests for continuous variables. All tests were two-tailed and a p-value <0.05 was considered statistically significant. Results A total of 236 neonates were included in the analysis. The mean maternal age was 26.3 ± 6.6 years, and the average gestational age at delivery was 29.4 ± 2.0 weeks. The mean birth weight percentile was 45.1 ± 26.8, and 23.8 % of infants were categorized as extremely preterm (< 28 weeks). The mean 1-minute and 5-minute APGAR scores were 6.3 ± 2.0 and 8.3 ± 1.5, respectively. The average hospital stay was 43.0 ± 24.4 days, and the mean duration of NICU admission was 23 ± 20 days. Sixty-four percent of patients required conventional mechanical ventilation, 13 % high-frequency ventilation, and nearly three-quarters (70.2 %) needed supplemental oxygen at 36 weeks postmenstrual age. Infectious complications were frequent, affecting 27.5 % of the population, while bacteremia and pneumonia were observed in 8.1 % and 4.2 %, respectively. The overall in-hospital mortality rate was 11.0 %. These data are summarized in Table 1 . The latent class analysis identified the 3-class model as the best-fitting structure compared with the 2-class alternative (AIC = 12 002, BIC = 12 440, aBIC = 12 141 vs. AIC = 12 541, BIC = 12 871, aBIC = 12 630), with significant Vuong–Lo–Mendell–Rubin and bootstrap likelihood ratio tests ( p < 0.01). Although the entropy value decreased slightly (0.83 vs. 0.86), the 3-class model provided a clinically coherent gradient of disease severity, consistent with the natural spectrum of bronchopulmonary dysplasia (BPD). The 2-class model captured only a binary contrast between mild–moderate and severe phenotypes, whereas the 3-class model discriminated a transitional group with intermediate morbidity. Table 2 summarizes the model fit indices. Mean posterior probabilities of class membership were high across all three latent classes (0.92, 0.89, and 0.91, respectively), confirming strong within-class homogeneity and minimal cross-assignment. Overlap was marginal between the first and second classes (≈ 0.06–0.07), suggesting a continuum of respiratory severity rather than discrete boundaries. Entropy values above 0.80 indicated reliable classification. These results are presented in Table 3 . When the clinical and demographic characteristics were compared among the three latent classes, the model delineated three distinct phenotypes that corresponded to progressively worsening pulmonary and systemic involvement ( Table 4 ). Class 1, which represented approximately two-thirds of the cohort, comprised more mature preterm infants with an average gestational age of around 30–32 weeks. These infants exhibited higher APGAR scores (6.8 ± 1.6 and 8.7 ± 1.3 at 1 and 5 minutes, respectively), shorter hospital stays, minimal need for invasive ventilation, and lower oxygen dependency. Only 2.5 % were extremely preterm, and mortality in this class was 5 %. Clinically, these patients exhibited a self-limited form of neonatal respiratory distress that resolved with supportive care, corresponding to a low-risk or recovering BPD phenotype. Class 2, accounting for about one-quarter of the cohort, represented a transitional phenotype with intermediate risk. These infants were typically born at 28–30 weeks of gestation, had lower APGAR scores (5.9 ± 2.1 at 1 minute and 8.1 ± 1.4 at 5 minutes), and required prolonged oxygen therapy and hospitalization. The mean scaled indicators for hospital and NICU stay (407 and 402, respectively) were nearly double those of Class 1, suggesting sustained respiratory instability and recurrent exacerbations. Bronchopulmonary dysplasia was present in 21.4 % of infants within this group, although mortality remained low (5 %). This phenotype represents intermediate respiratory morbidity, characterized by prolonged but ultimately reversible lung disease with potential long-term pulmonary vulnerability. Class 3 encompassed only 8 % of the neonates but represented the most severe and clinically unstable phenotype. Virtually all infants in this class were extremely preterm (94 %), with markedly low APGAR scores (4.5 ± 2.3 at 1 minute and 6.6 ± 2.1 at 5 minutes), extensive exposure to mechanical ventilation, and near-universal oxygen dependence. Despite intensive support, mortality reached 88 %. The low average length of hospital stay in this class contrasted with their critical condition, reflecting early fatal outcomes rather than clinical recovery. These patients exemplify a high-risk or non-surviving phenotype, characterized by severe pulmonary immaturity, multiorgan dysfunction, and limited capacity for postnatal lung repair. Comparative statistical analyses confirmed significant differences in clinical outcomes among the three latent classes. The Kruskal–Wallis test showed highly significant differences for both total hospital stay (χ²(2) = 109.8, p < 0.001) and duration of NICU stay (χ²(2) = 93.3, p < 0.001). Infants classified within the intermediate-risk group (Class 2) had the longest hospital and NICU stays, consistent with prolonged survival and persistent respiratory support needs, while those in Class 3 had shorter stays due to early death. Mortality also differed markedly between classes (Pearson χ²(2) = 110.9, p < 0.001). Survival was high in Class 1 (95 %) and Class 2 (95 %) but dramatically lower in Class 3 (12 %). The combined results delineate a clear, statistically robust gradient in disease severity, with a stepwise increase in ventilatory support, oxygen dependency, and mortality across classes. Clinically, these findings confirm that the latent structure reflects the biological continuum of bronchopulmonary dysplasia, integrating perinatal maturity, respiratory adaptation, and systemic complications into a unified disease spectrum. The progression from Class 1 to Class 3 mirrors the transition from mild, self-limiting pulmonary dysfunction to severe, irreversible lung disease associated with high lethality. Importantly, Class 2 captures a clinically relevant intermediate profile with potential for survival but risk of chronic respiratory sequelae. The latent class approach therefore provides a robust and empirically grounded framework for phenotyping preterm infants with heterogeneous respiratory outcomes, enabling more precise prognostic modeling and risk stratification in neonatal pulmonary medicine. Discussion. Our latent class analysis identified three clinically distinct subgroups of preterm infants with progressive gradients in gestational age, respiratory support, and mortality, consistent with the evolution of conceptual frameworks that have redefined bronchopulmonary dysplasia (BPD) over the last decade. Early in this paradigm shift, Whang proposed that BPD should no longer be viewed as a single disease entity but rather as a heterogeneous syndrome encompassing multiple pulmonary phenotypes arising from different pathways of injury and repair (11). He described four major phenotypes—parenchymal, interstitial, airway, and vascular—each reflecting the predominant compartment of injury in the immature lung and determined by a combination of antenatal and postnatal insults. Within this model, our three-class solution parallels Bhandari’s initial phenotypic architecture: the first class corresponds to a mild airway or parenchymal phenotype characterized by limited postnatal inflammation and rapid recovery; the second class reflects a more severe parenchymal–interstitial pattern associated with prolonged ventilation and alveolar arrest; and the third class aligns with the vascular phenotype, marked by growth restriction, pulmonary hypertension, and high mortality. Building upon this framework, Pierro and colleagues (2022) expanded the concept by introducing the endotype–phenotype continuum of prematurity, in which two major antenatal endotypes—inflammatory/infectious and placental dysfunction—drive divergent pulmonary phenotypic expressions (12). In this context, our first latent class likely represents infants primarily affected by transient inflammatory endotypes with predominant postnatal recovery, whereas the third class corresponds to a vascular phenotype emerging from placental dysfunction, with profound pulmonary vascular maladaptation and early circulatory failure. The intermediate class appears to lie between these two biological trajectories, representing mixed injury patterns with sustained oxygen dependence but intermediate survival. More recently, Gilfillan and Bhandari (2023) refined this multidimensional classification by delineating five overlapping pulmonary phenotypes—parenchymal, interstitial, peripheral airway, central airway, and vascular—each associated with specific structural, molecular, and functional alterations in the developing lung (13). Within this modern schema, the first class in our analysis aligns with a mild peripheral airway or parenchymal phenotype; the second class corresponds to a mixed parenchymal–interstitial pattern characterized by alveolar simplification and fibroproliferative remodeling; and the third class mirrors the vascular phenotype, driven by impaired angiogenesis and placental-origin hypoxemia. The sequential concordance between our data-driven latent structure and these biologically grounded frameworks—from Bhandari’s mechanistic model to Pierro’s endotype-based taxonomy and Gilfillan’s multidimensional phenotypic refinement—reinforces the biological plausibility of our findings. Collectively, these results support the notion that BPD represents a spectrum of interrelated pulmonary diseases sharing developmental origins but diverging in pathophysiologic pathways and clinical outcomes. Integrating this mechanistic–phenotypic perspective into neonatal research may enhance prognostic precision, guide risk stratification, and ultimately enable personalized respiratory care for preterm infants. Wu and colleagues, in a single-center cohort of 76 extremely preterm infants (<32 weeks’ gestation) with severe bronchopulmonary dysplasia (BPD) evaluated at term-equivalent age, identified three major pathophysiologic components—parenchymal lung disease, pulmonary hypertension, and large airway malacia—whose coexistence defined distinct structural–functional phenotypes associated with increasing mortality and tracheostomy risk(14). In contrast, our latent class analysis, performed in a broader cohort of preterm infants from a middle-income setting, identified three clinically derived subgroups based on patterns of respiratory support, oxygen dependency, and survival, without relying on advanced imaging. Despite methodological differences, both studies converge conceptually: the progressive clinical severity observed in our three-class model mirrors the structural gradient described by Wu et al., in which the vascular or mixed vascular–parenchymal phenotype represents the most severe expression of disease. Differences in phenotypic distribution likely stem from the inclusion of milder BPD cases in our cohort, the use of unsupervised statistical modeling rather than radiologic classification, and contextual disparities such as higher prevalence of fetal growth restriction, infection-related prematurity, and variability in neonatal intensive care resources in our population. Taken together, these findings support the notion that BPD represents a spectrum of pulmonary injury phenotypes that can be captured either through imaging-based structural markers or through data-driven latent class approaches, both converging toward a unified model of developmental lung disease with shared biological underpinnings but divergent clinical trajectories. The recent work by McAnany et al. (2025) strengthens the evidence that phenotype-based approaches can improve the prediction of outcomes in infants with severe bronchopulmonary dysplasia (BPD)(15). In a cohort of 100 infants discharged from a tertiary neonatal center, the authors showed that the Neonatal Research Network and BPD Collaborative definitions outperformed the classic 2001 NIH criteria in predicting tracheostomy and death within one year (p < 0.001), with large airway disease emerging as the phenotype most strongly associated with poor outcomes (OR 10.5; 95% CI: 1.6–68.1). Compared with our latent class model, which identified three clinical subgroups along a gradient of maturity, oxygen dependency, and mortality, their anatomically defined phenotypes provide complementary evidence that the structural compartment involved—airway, parenchymal, or vascular—determines prognosis. Both studies converge on the vascular phenotype as the most severe trajectory, reflecting placental dysfunction and pulmonary vascular maladaptation. Differences in case severity, inclusion criteria, and diagnostic resources likely explain the absence of a discrete large-airway phenotype in our cohort, yet the alignment in pathophysiologic patterns reinforces that data-driven and structural frameworks capture the same biological heterogeneity of BPD and support phenotype-based strategies for individualized neonatal care. The study by Zhang et al. (2025) offers a valuable contribution to understanding how environmental and physiological factors at high altitude influence the development and severity of bronchopulmonary dysplasia (BPD) in preterm infants, while also introducing a machine learning–based risk prediction framework(16). Conducted at 1500 m above sea level, this retrospective matched cohort of 378 infants identified maternal hypertension, initial FiO₂ > 30%, and prolonged invasive ventilation as the strongest predictors of BPD, with the XGBoost model achieving an AUC = 0.89 and F1 = 0.82, outperforming traditional logistic regression approaches. Compared with our latent class analysis, which identified three clinical subtypes driven by gestational maturity, oxygen dependency, and mortality, Zhang’s study focused on environment-modified risk hierarchies rather than latent phenotypes. The distinct predominance of early respiratory and vascular factors at altitude—reflecting chronic hypoxemia and impaired pulmonary angiogenesis—parallels the vascular phenotype identified in our cohort, supporting the concept that BPD phenotypic expression is context-sensitive and environmentally modulated. Methodological contrasts, such as Zhang’s reliance on algorithmic prediction and exclusion of extreme growth restriction, versus our population-based inclusion and data-driven phenotyping, may explain differences in phenotype prevalence but converge on a shared conclusion: early hypoxic stress, maternal vascular dysfunction, and prolonged ventilatory exposure are central, cross-context determinants of severe BPD trajectories. These complementary insights underscore the need to integrate altitude-specific physiology and data-driven phenotypic frameworks to refine risk stratification and guide individualized neonatal care across diverse settings. This study has several limitations that should be acknowledged. First, its retrospective and single-center design may limit generalizability, as clinical practices, availability of neonatal resources, and thresholds for respiratory support can vary across institutions and health systems. Second, the analysis relied exclusively on clinical and perinatal data, without inclusion of advanced imaging, echocardiographic, or biomarker parameters that could refine phenotypic characterization. Third, the classification of latent classes was derived from routinely collected variables, which—although clinically meaningful—may not fully capture the structural heterogeneity of BPD described in recent mechanistic and imaging-based studies. Additionally, the absence of long-term follow-up precludes assessment of post-discharge outcomes such as growth, rehospitalization, or neurodevelopment. Finally, given the limited sample size of extremely preterm infants, some subgroup estimates, particularly within the most severe phenotype, should be interpreted with caution. Conclusion In this cohort of preterm infants born at ≤32 weeks’ gestation, latent class analysis identified three distinct phenotypic subgroups characterized by progressive gradients in maturity, respiratory support, and survival. These classes mirror the emerging biologic and structural phenotypes described in recent literature—ranging from mild inflammatory and parenchymal forms to severe vascular phenotypes associated with growth restriction and high mortality. Our findings support the concept that bronchopulmonary dysplasia represents a continuum of developmental lung disorders rather than a single disease entity and demonstrate that data-driven phenotyping can uncover clinically and biologically meaningful heterogeneity even in resource-limited settings. Integrating such latent structures with imaging, molecular, and longitudinal data may advance the transition toward precision neonatal medicine, enabling individualized risk stratification and targeted interventions for infants with evolving BPD. Table 1. Descriptive statistics for the neonatal population (n = 236) Variable Mean ± SD Maternal age (years) 26.3 ± 6.6 Gestational age by ultrasound (weeks) 29.4 ± 2.0 Birth weight percentile 45.1 ± 26.8 Length percentile 36.0 ± 26.7 Percentile at 36 weeks 21.0 ± 19.1 1-minute APGAR 6.3 ± 2.0 5-minute APGAR 8.3 ± 1.5 Prenatal controls 5.0 ± 2.9 Duration of rupture of membranes (hours) 30.7 ± 99.9 Hospital stay (days) 43.0 ± 24.4 NICU stay (days) 23.4 ± 20.2 Respiratory support (mean ± SD or %) Variable % or mean ± SD Invasive ventilation (days) 6.8 ± 12.1 Oxygen therapy (days) 39.4 ± 26.2 Conventional ventilation used 64.0 % High-frequency ventilation used 13.1 % CPAP used 46.0 % VMNI used 58.3 % CNAF used 28.9 % CNS used 86.4 % Oxygen at 36 weeks 70.2 % Perinatal and maternal characteristics (%) Variable % or mean ± SD Male sex 51.9 % Urban residence 25.2 % Maternal smoking 0.4 % Premature extreme (<28 weeks) 23.8 % Fetal growth restriction 27.7 % Suspected intra-amniotic infection 8.1 % Cause of prematurity (recorded) 63.6 % Use of antibiotics 26.7 % Morbidity and complications (%) Condition % Catheter-related bacteremia 8.1 % Ventilator-associated pneumonia (NAVM) 6.4 % Pneumonia (clinical) 4.2 % Bacteremia (isolated) 8.1 % Urinary tract infection 7.2 % Necrotizing enterocolitis 1.7 % Retinopathy of prematurity 1.3 % Ductus arteriosus persistence (36 w) 14.4 % Pulmonary hypertension 4.2 % HIV infection 5.5 % Apneas 68.2 % Outcomes (%) Variable % Discharged before 36 weeks 50.0 % Hospital death 11.0 % DBP-related fatality 2.5 % All-cause death 11.0 % Table 2. Latent class analysis fit statistics for the neonatal population (n = 236) 2 12,541 12,871 12,63 <.01 <.01 0.86 0.91 – 0.94 3 12,002 12,44 12,141 <.01 <.01 0.83 0.88 – 0.92 Table 3. Mean latent class probabilities for the 3-class solution of the neonatal 1 – Mild/low-risk 0.922 0.043 0.035 2 – Moderate/intermediate-risk 0.039 0.916 0.045 3 – Severe/high-risk 0.037 0.042 0.921 Table 4. Descriptive statistics of demographic and clinical characteristics by class for the 3-class solution in the neonatal population (n = 236) Sex (male), % 50.0 56.1 41.2 Extreme prematurity (<28 w), % 2.5 37.8 94.1 Fetal growth restriction, % 28.3 28.6 17.6 1-min APGAR, mean (SD) 6.8 (1.6) 5.9 (2.1) 4.5 (2.3) 5-min APGAR, mean (SD) 8.7 (1.3) 8.1 (1.4) 6.6 (2.1) Gestational age (weeks, scaled) 386.7 217.3 100.0 Length of stay (scaled) 228.3 407.1 147.1 NICU stay (scaled) 220.8 402.0 200.0 Days on oxygen (scaled) 223.3 414.3 147.1 Conventional ventilation, % 71.7 55.6 58.8 Death, % 5.0 5.1 88.2 References 1. Jobe AH, Bancalari E. Bronchopulmonary dysplasia. Am J Respir Crit Care Med. 2001;163(7):1723-9.2. Stoll BJ, Hansen NI, Bell EF, Walsh MC, Carlo WA, Shankaran S, et al. Trends in Care Practices, Morbidity, and Mortality of Extremely Preterm Neonates, 1993-2012. JAMA. 2015;314(10):1039-51.3. Thebaud B, Goss KN, Laughon M, Whitsett JA, Abman SH, Steinhorn RH, et al. Bronchopulmonary dysplasia. Nat Rev Dis Primers. 2019;5(1):78.4. Berlin KS, Williams NA, Parra GR. An introduction to latent variable mixture modeling (part 1): overview and cross-sectional latent class and latent profile analyses. J Pediatr Psychol. 2014;39(2):174-87.5. Momany AM, Jasper E, Markon KE, Nikolas MA, Ryckman KK. Latent class analysis to characterize neonatal risk for neurodevelopmental differences. J Child Psychol Psychiatry. 2023;64(1):100-9.6. Hendryx M, Chojenta C, Byles JE. Latent Class Analysis of Low Birth Weight and Preterm Delivery among Australian Women. J Pediatr. 2020;218:42-8 e1.7. Collaco JM, Aoyama BC, Rice JL, McGrath-Morrow SA. Influences of environmental exposures on preterm lung disease. Expert Rev Respir Med. 2021;15(10):1271-9.8. Xu YP, Chen Z, Dorazio RM, Bai GN, Du LZ, Shi LP. Risk factors for bronchopulmonary dysplasia infants with respiratory score greater than four: a multi-center, prospective, longitudinal cohort study in China. Sci Rep. 2023;13(1):17868.9. Swarray-Deen A, Sepenu P, Mensah TE, Osei-Agyapong J, Sefogah PE, Appiah-Sakyi K, et al. Preterm birth in low-middle income Countries. Best Pract Res Clin Obstet Gynaecol. 2024;95:102518.10. Buendia JA, Ramirez Velasquez C, Benjumea-Bedoya D. Bronchopulmonary dysplasia: Incidence and severity in premature infants born at high altitude. Pediatr Pulmonol. 2022;57(2):470-5.11. Wang SH, Tsao PN. Phenotypes of Bronchopulmonary Dysplasia. Int J Mol Sci. 2020;21(17).12. Pierro M, Van Mechelen K, van Westering-Kroon E, Villamor-Martinez E, Villamor E. Endotypes of Prematurity and Phenotypes of Bronchopulmonary Dysplasia: Toward Personalized Neonatology. J Pers Med. 2022;12(5).13. Gilfillan M, Bhandari V. Pulmonary phenotypes of bronchopulmonary dysplasia in the preterm infant. Semin Perinatol. 2023;47(6):151810.14. Wu KY, Jensen EA, White AM, Wang Y, Biko DM, Nilan K, et al. Characterization of Disease Phenotype in Very Preterm Infants with Severe Bronchopulmonary Dysplasia. Am J Respir Crit Care Med. 2020;201(11):1398-406.15. McAnany S, Akangire G, Sherman A, Sampath V, Manimtim W. Predicting Clinical Outcomes of Severe Bronchopulmonary Dysplasia through New Definitions and Phenotypes. Am J Perinatol. 2025.16. Zhang H, Wang F, Jiang O, Lin Y, Tang L, Li Z, et al. Machine learning-based risk prediction models for bronchopulmonary dysplasia in preterm infants: a high-altitude cohort study. BMJ Paediatr Open. 2025;9(1). Information & Authors Information Version history V1 Version 1 11 November 2025 Copyright This work is licensed under a Non Exclusive No Reuse License. Keywords bronchopulmonary dysplasia colombia latent class analysis phenotypes preterm Authors Affiliations Andrea Parra Buitrago 0009-0002-9665-5791 [email protected] Universidad Pontificia Bolivariana Clinica Universitaria Bolivariana View all articles by this author Andrea Jaramillo Cerezo Universidad Pontificia Bolivariana Clinica Universitaria Bolivariana View all articles by this author Jefferson Buendia 0000-0003-2404-6612 Universidad de Antioquia Facultad de Medicina View all articles by this author Metrics & Citations Metrics Article Usage 236 views 172 downloads .FvxKWukQNSOunydq8rnd { width: 100px; } Citations Download citation Andrea Parra Buitrago, Andrea Jaramillo Cerezo, Jefferson Buendia. Uncovering Hidden Phenotypes of Bronchopulmonary Dysplasia: A Latent Class Analysis of Preterm Infants in a Middle-Income Setting. Authorea . 11 November 2025. DOI: https://doi.org/10.22541/au.176289131.11095672/v1 If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download. For more information or tips please see 'Downloading to a citation manager' in the Help menu . Format Please select one from the list RIS (ProCite, Reference Manager) EndNote BibTex Medlars RefWorks Direct import Tips for downloading citations document.getElementById('citMgrHelpLink').addEventListener('click', function() { popupHelp(this.href); return false; }); $(".js__slcInclude").on("change", function(e){ if ($(this).val() == 'refworks') $('#direct').prop("checked", false); $('#direct').prop("disabled", ($(this).val() == 'refworks')); }); View Options View options PDF View PDF Figures Tables Media Share Share Share article link Copy Link Copied! Copying failed. Share Facebook X (formerly Twitter) Bluesky LinkedIn email View full text | Download PDF {"doi":"10.22541/au.176289131.11095672/v1","type":"Article"} Now Reading: Share Figures Tables Close figure viewer Back to article Figure title goes here Change zoom level Go to figure location within the article Download figure Toggle share panel Toggle share panel Share Toggle information panel Toggle information panel Go to previous graphic Go to next graphic Go to previous table Go to next table All figures All tables View all material View all material xrefBack.goTo xrefBack.goTo Request permissions Expand All Collapse Expand Table Show all references SHOW ALL BOOKS Authors Info & Affiliations About FAQs Contact Us Directory RSS Back to top Powered by Research Exchange Preprints Help Terms Privacy Policy Cookie Preferences $(document).ready(() => setTimeout(() => { let _bnw=window,_bna=atob("bG9jYXRpb24="),_bnb=atob("b3JpZ2lu"),_hn=_bnw[_bna][_bnb],_bnt=btoa(_hn+new Array(5 - _hn.length % 4).join(" ")); $.get("/resource/lodash?t="+_bnt); },4000)); (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'9fe975e34ff81640',t:'MTc3OTI2MDM4NQ=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();