Exploring the potential effects of acute respiratory failure on patients with early-stage bronchopulmonary cancer Authors: Ziwen Lv, Qi Feng, Junhui Gu, Zhuo Gao, Dong Liu Published: 2025 ABSTRACT AbstractAcute respiratory failure is a common disease, impacting various cancer survivors and reflecting post-surgery complications. However, little evidence based on respiratory failure has been developed for prognostic assessment and prediction in patients with bronchopulmonary cancer. Data from the Medical Information Mart for Intensive Care-IV (MIMIC-IV v2.2) database were analyzed to identify risk factors associated with acute respiratory failure in patients with bronchopulmonary cancer. The primary outcome was the occurrence of acute respiratory failure. Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) regression model, and significant predictors were incorporated into a nomogram for individualized risk prediction. Among the 1,318 included patients with bronchopulmonary cancer, 35.6% developed acute respiratory failure. Ten independent risk factors were identified and used to construct the nomogram: activated partial thromboplastin time (PTT), glucose, blood urea nitrogen (BUN), neutrophil percentage, sodium, respiratory rate (RR), heart rate (HR), chronic obstructive pulmonary disease (COPD), shock, and race. The model exhibited good discrimination, with an area under the ROC curve of 0.741. Internal validation using 1000 bootstrap replicates demonstrated strong agreement between the calibration curve and the ideal reference, indicating satisfactory model performance. We developed and validated nomograms for patients with early-stage bronchopulmonary cancer, which could provide individual prediction of acute respiratory failure.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-24267-2.Keywords: Bronchopulmonary cancer, Respiratory failure, MIMIC-IV database, PrognosisSubject terms: Diseases, Medical research, Risk factors Introduction Bronchopulmonary cancer represents a significant global health burden, accounting for substantial annual morbidity and mortality1. Sepsis (44%) and acute respiratory failure (31%) have been identified as the primary reasons for ICU admission among these patients2. Acute respiratory failure (RF), in particular, is a critical complication associated with high mortality in bronchopulmonary cancer3. Its pathogenesis often involves airway obstruction due to tumor progression, resulting in impaired ventilation. Additionally, pulmonary infections—common in these patients—can exacerbate ventilatory dysfunction and precipitate RF4. Since respiratory failure symptoms may be obscured by those of the underlying malignancy, early recognition and intervention are essential. Nomograms have emerged as valuable predictive tools in oncology, enabling individualized risk estimation for various clinical outcomes5–8. Evidence suggests that such models can outperform clinical judgment alone9,10. This study aims to develop and validate a nomogram for predicting the risk of respiratory failure in patients with bronchopulmonary cancer. The model may facilitate early identification of high-risk individuals, allowing timely intervention, personalized treatment selection, and ultimately, improved clinical outcomes. Materials and methods Study population This is a retrospective observational cohort study with longitudinal follow-upof patients. Medical Information Mart for Intensive Care-IV (MIMIC-IV-2.2) is a freely accessible data. The IRB has approved the waiver of informed consent and the data-sharing initiative. All private information in the database repository has been removed, rendering informed consent unnecessary for data collection. We have completed the Collaborative Institutional Training Initiative (CITI) course and passed the exams for “Conflict of Interest” and “Data or Specimen Only Research” (ID: 55,109,354). All methods were performed in accordance with the relevant guidelines and regulations. One author (Ziwen Lv) has become a credentialed used on PhysioNet and that they have signed the data use agreement11,12. Patients diagnosed with early bronchopulmonary cancer and admitted to the ICU were included according to the International Classification of Diseases (supplementary material S1).Early-stage bronchopulmonary carcinoma is defined according to the 8th edition TNM staging criteria of the American Joint Committee on Cancer (AJCC) as patients diagnosed with carcinoma in situ (Tis), stage I, or stage II disease.Exclusion criteria were as follows:(1) age < 18 years; (2) not first admission (3) missing data on patient variables > 30%(Fig. 1). Flowchart of study patients. ARF acute respiratory failure, MIMIC Medical Information Mart for Intensive Care, ICU intensive care unit. Data extraction In this study, Navicat Premium software (version 16) was used to extract demographic information, laboratory indicators, and comorbidities from the MIMIC-IV database by executing Structured Query Language (SQL). Demographic information included gender, age, race. Vital signs, respiratory rate, heart rate, systolic blood pressure, diastolic blood pressure at first admission. Laboratory indicators include serum albumin, red blood cells (RBC), white blood cells (WBC), red blood cell distribution width (RDW), hematocrit, hemoglobin, lactate, international normalized ratio (INR), serum potassium, serum sodium, serum total calcium, serum chloride, platelets, blood glucose, blood creatinine, Neutrophil percentage, lymphocyte percentage, prothrombin time (PT), activated partial thromboplastin time (PTT), anion gap (AG), blood urea nitrogen (BUN), Peripheral Capillary Oxygen Saturation, pulse oxygen saturation, PH. Patients’ comorbidities and personal history were determined according to ICD-9 and ICD-10, including chronic obstructive pulmonary disease, and shock (supplementary material S1). For laboratory indicators measured multiple times after admission, the first measurement was used in this study. To minimize reverse causation bias, data collected after the outcome event were considered invalid. Missing values for laboratory indicators are common in the MIMIC-IV database. Because missing data may weaken the robustness and validity of the study’s conclusions, multiple interpolation was used to deal with these missing data13. For variables with more than 30% missing data, exclude them; for those with missing data not exceeding 30%, apply multiple imputation to improve the data 14.The interpolation model was set to perform 5 iterations, generating 5 complete datasets. Both the outcome variable and all analytical variables were incorporated into the interpolation process to maintain unbiased results. The final model parameters represent the combined results of these 5 datasets. Primary endpoint The primary outcome of this study was acute respiratory failure in patients with early-stage bronchopulmonary cancer. Statistical analysis Statistical analysis was performed using R version 4.2.3 and SPSS version 26.0 software. A normality test was performed at the preliminary stage of data analysis. The normality of continuous variables was assessed using the Kolmogorov–Smirnov test. For normally distributed continuous variables, mean + /- SD was reported; for non-normally distributed continuous variables, median (IQR); and for categorical variables, count (%). In analyzing baseline characteristics, the Kruskal–Wallis rank sum test was used for continuous variables and the Fisher exact probability test was used for count variables with theoretical counts less than 10. Based on clinical knowledge and literature reports, we initially selected a set of predictor variables potentially associated with outcomes. Subsequently, the aforementioned LASSO regression was employed to screen out predictors with non-zero coefficients from these variables. LASSO regression was primarily chosen due to its suitability for variable selection in high-dimensional data and its ability to effectively address multicollinearity among predictors through the L1 regularization penalty function. λ Value Selection: The optimal λ (lambda) value was selected via tenfold cross-validation. Specifically, the λ value corresponding to the 1-standard-error rule—which minimizes the cross-validation error—was chosen. This λ value ensures model parsimony while achieving optimal predictive performance. Finally, all variables selected by LASSO were incorporated into a multivariate logistic regression model to simultaneously assess the independent influence of each variable. The performance of the Nomograms was evaluated using the AUC of ROC curves., and the accuracy of the model was assessed by resampling 1000 times and plotting calibration curves. We used Bootstrap self-help sampling method for model overfitting correction, raw calibration curve: based on the initial model prediction of the raw data, bias correction curve: average of 1000 Bootstrap biases by translating the raw curve downwards. Finally, decision curve analysis (DCA) was used to evaluate the validity and clinical impact of the model. Differences were statistically significant when P values < 0.05 were defined. Results Baseline characteristics of patients As per the inclusion and exclusion criteria, a total of 1318 patients with bronchopulmonary cancer were included in this study from the MIMIC -IV database. Acute respiratory failure was present in 35.6% of these patients. Table 1 compares acute respiratory failure with no acute respiratory failure and finds that patients in the respiratory failure group had higher respiratory rate, heart rate, WBC, serum sodium, serum potassium, serum total calcium, RDW, neutrophil percentage, INR, blood urea nitrogen, blood creatinine, blood glucose, PT, PTT, and anion gap; lower hemoglobin and Spo2; higher prevalence of respiratory failure combination with shock, chronic obstructive pulmonary disease. Table 1Baseline characteristics of the ARF and non-ARF group.CharacteristicNon-ARF(64.4%)ARF(35.6%)P-valueAge67.9 ± 11.167.6 ± 10.10.612Weight (Kg)73.4 ± 18.872.4 ± 19.70.384HR (Beat)114.0 ± 24.3126.9 ± 24.7 < 0.001RR (Beat)31.4 ± 10.936.3 ± 9.8 < 0.001Systolic pressure (mmhg)160.5 ± 335.1182.1 ± 531.10.368Diastolic pressure(mmhg)242.1 ± 3711.9115.6 ± 25.60.461RBC (m/uL)4.0 ± 0.63.9 ± 0.70.064Lactate (mmol/L)2.2 ± 1.82.1 ± 1.70.924WBC (K/uL)15.6 ± 9.118.3 ± 10.6 < 0.001Sodium (mEq/L)140.7 ± 4.2142.6 ± 5.1 < 0.001Platelets (K/uL)332.1 ± 154.3343.0 ± 178.00.244Total Calcium (mg/dL)9.1 ± 0.89.2 ± 1.10.043Hemoglobin (g/L)11.7 ± 1.911.4 ± 2.00.003RDW (%)15.8 ± 2.316.4 ± 2.6 < 0.001Lymphocytes%15.1 ± 15.314.3 ± 16.00.336Potassium (mEq/L)4.8 ± 0.85.0 ± 0.8 < 0.001Neutrophils%82.0 ± 11.585.0 ± 10.6 < 0.001INR1.5 ± 1.11.7 ± 1.10.012HCT(%)35.8 ± 5.435.5 ± 5.90.344Chloride(mEq/L)105.7 ± 5.0106.1 ± 5.90.190BUN (mg/dL)28.1 ± 19.436.6 ± 27.1 < 0.001Glucose (mmol/l)176.5 ± 83.8201.0 ± 89.3 < 0.001creatinine (mg/dL)1.3 ± 1.11.5 ± 1.20.001PT (s)16.5 ± 10.818.2 ± 11.90.008PTT45.1 ± 32.555.6 ± 40.6 < 0.001AG (mEq/L)16.9 ± 3.717.8 ± 4.3 < 0.001PH7.4 ± 0.17.4 ± 0.10.002Spo2 (%)87.8 ± 5,287.2 ± 5.90.001po2 (%)47.9 ± 15.059.6 ± 21.2 < 0.001Sex (man)400 (47.1%)240 (51.2%)0.158Race (white)633 (74.6%)310 (66.1%)0.004COPD100 (11.8%)104 (22.2%) < 0.001DM200 (23.6%)129 (27.5%)0.113Shock63 (7.4%)113 (24.1%) < 0.001ARF acute respiratory failure, HR: heart rate, RR respiratory rate, RBC red blood cell, WBC white blood Cell, RDW red blood cell distribution width, INR international normalized ratio,PT prothrombin time, PTT activated partial thromboplastin time, AG anion gap, HCT hematocrit, BUN blood urea nitrogen, AG anion gap, COPD chronic obstructive pulmonary disease, DM diabetes. P-value less than 0.05 is expressed in bold. Baseline characteristics of the ARF and non-ARF group. ARF acute respiratory failure, HR: heart rate, RR respiratory rate, RBC red blood cell, WBC white blood Cell, RDW red blood cell distribution width, INR international normalized ratio, PT prothrombin time, PTT activated partial thromboplastin time, AG anion gap, HCT hematocrit, BUN blood urea nitrogen, AG anion gap, COPD chronic obstructive pulmonary disease, DM diabetes. P-value less than 0.05 is expressed in bold. Selection of risk factors for acute respiratory failure in bronchopulmonary cancer patients by LASSO modeling In this study, the occurrence of acute respiratory failure in bronchopulmonary cancer patients who were admitted to the ICU was used as the dependent variable, and the LASSO regression algorithm was utilized to screen the variables and select the optimal λ value through tenfold cross-validation Figs. 2 and 3. The minimum number of variables was included to obtain the most streamlined predictive model while ensuring goodness of fit. The LASSO regression screening variables included PTT, Glucose, BUN, Neutrophil percentage, Sodium, Respiratory rate, Heart rate,COPD, Shock, and Race Table 2 Log(λ) and the error model. Lambda is the regularisation parameter. The 2 dashed lines in Fig. 2 indicate lambda.min and lambda.1se, respectively. The X-axis represents the increase in penalty strength (log(λ)), and the Y-axis represents the mean squared error (MSE) of prediction of the cross-validation estimate, with lower error representing better model fit. The curves show the trend of the prediction error with the penalty strength: the model is complex and may be overfitted when the penalty is weak (left); the model is too simple leading to underfitting when the penalty is strong (right); and there is a region of minimum error in the middle. The vertical dashed line λ.min identifies the λ corresponding to the absolute minimum of cross-validation error, representing the theoretically best fit goodness of fit. The vertical dashed line λ.1se identifies the λ corresponding to the point where the error value rises to min(MSE_cv) + 1SE. at the λ.1se dashed line, the model uses fewer variables than at λ.min.Lambda.min denoted the value of λ when the model error was minimal. Lambda.1se denoted the model error within a standard error range of λ. Log(λ) and the LASSO regression coefficient. The X-axis is the logarithm of the penalty coefficient λ (log(λ)) and the Y-axis is the standardised regression coefficient. Each coloured path represents the coefficient change of a predictor variable; the vertical dashed line (λ.1se) indicates the optimal value of λ selected by ten-fold cross-validation (based on the cross-validation error minimisation principle of Fig. 2); coefficients of the variables to the right of this dashed line are compressed to 0. Table 2Multivariable LASSO-logistic regression analysis of bronchopulmonary cancer.Risk factorsOdds ratio95%CIP-valueRace (white)0.9020.579–1.4040.046HR1.0211.016–1.026 < 0.0001RR1.0561.044–1.0740.006Neutrophils1.0271.015–1.0390.012BUN1.0161.011–1.0210.042Glucose1.0021.001–1.0030.043PTT1.0081.005–1.0110.035Sodium1.0951.067–1.1230.001Shock3.9602.840–5.523 < 0.0001COPD2.1341.578–2.884 < 0.0001HR heart rate, RR respiratory rate, PTT activated partial thromboplastin time, AG anion gap, HCT hematocrit, BUN blood urea nitrogen, COPD chronic obstructive pulmonary disease. OR Odds Ratio; P-value less than 0.05 is expressed in bold. Multivariable LASSO-logistic regression analysis of bronchopulmonary cancer. HR heart rate, RR respiratory rate, PTT activated partial thromboplastin time, AG anion gap, HCT hematocrit, BUN blood urea nitrogen, COPD chronic obstructive pulmonary disease. OR Odds Ratio; P-value less than 0.05 is expressed in bold. Construction of the prediction model Based on the results of LASSO regression, a clinical prediction model for acute respiratory failure in bronchopulmonary cancer patients was constructed (Fig. 4). Nomograms are a visual representation of complex regression equations that score each patient’s identified risk factors or clinical characteristics. The level of influence of the risk factors in the model on the likelihood of an acute respiratory failure outcome was determined by assigning a score to the level of each factor on a scale of 0 to 100 points on Nomograms, and the scores for each factor were summed to obtain a Total Points score. The red dots mark the specific score locations for the example patients in this section. The total score is mapped to the ARF probability via a logistic transformation formula. The predicted probability of an individual ending event was calculated based on the functional relationship that exists between the total score and the probability of occurrence of the ending event. The results of one case showed that the sum of scores was 260, and the risk of acute respiratory failure during the patient’s stay in the ICU was approximately 86.3%. The 86.3% indicates the predicted probability value for this patient. Nomogram of acute respiratory failure occurs in patients with early bronchopulmonary cancer. Validation of the model The area under the curve(AUC) of the prediction model = 0.741, P-value 0.001 (Fig. 5A) ; The horizontal coordinate of the calibration curve is the predicted probability of the model, and the vertical axis indicates the actual probability of acute respiratory failure in patients with early-stage bronchopulmonary cancer. The apparent curve represents the internal correction curve, the Bais-corrected curve represents the bias correction by 1000 times repeated self-help sampling, and the calibration prediction curve is consistent with the ideal curve as seen in the figure (Fig. 5B) , suggesting that the accuracy of the model can be. Based on the clinical feasibility of the intervention strategies described above, we assessed the net benefit of model-guided interventions through decision curve analysis (Fig. 6) . The vertical coordinate of the DCA curve in Figure represents the net benefit rate, which can be interpreted as the difference between the benefit of intervening on true positives and the cost of intervening on false positives. The horizontal coordinate represents the threshold probability, which corresponds to the model’s probability of detecting an acute respiratory failure outcome. The black line therein is the benefit curve that represents the full population with no intervention at all; the net benefit of no intervention must be zero. The purple line is an indication that if everyone receives the intervention,then the net benefit decreases as the threshold probability increases. The red curve represents the benefit curve of a complex model for the prediction of acute respiratory failure in patients with early-stage bronchopulmonary cancer by a logistic regression model with the inclusion of all the independent risk factors. It is higher than both extreme curves, the model is considered better.Our model demonstrated a higher net benefit compared to the all-intervention or no-intervention strategies. The receiver operating characteristic curve (ROC) and Standard curve of model. (A) ROC is used to predict acute respiratory failure occurs in patients with early bronchopulmonary cancer; (B)Standard curve for clinical prediction model for development of acute respiratory failure occurs in patients with early bronchopulmonary cancer. The horizontal coordinate of the calibration curve is the prediction probability of the model, and the vertical coordinate is the probability of the observation in the actual data.The ideal curve is an ideal prediction where the predicted probability is exactly the same as the actual probability, the Apparent curve represents the internal correction curve, and the Bais-corrected curve represents the bias correction by 1000 repetitions of self-help sampling.The calibration curve reflects the consistency between the model’s predicted risk and the actual observed risk. The calibration curve of this model closely aligns with the ideal curve, indicating that the model’s predictive accuracy is excellent. Decision-analysis curves (DCA) of nomogram. DCA of a clinical prediction model for the development of acute respiratory failure occurs in patients with early bronchopulmonary cancer. Discussion The occurrence of acute respiratory failure in patients with bronchopulmonary cancer is multifactorial, underscoring the importance of identifying independent risk factors. We developed a nomogram to predict acute respiratory failure in ICU-admitted bronchopulmonary cancer patients based on the following predictors: PTT, glucose, BUN, neutrophil percentage, sodium, respiratory rate, heart rate, COPD, shock, and race. Neutrophil-mediated lung injury is thought to occur through several mechanisms. Elevated pro-inflammatory mediators promote neutrophil recruitment and sequestration in pulmonary capillaries, where they contribute to tissue damage via protease release, reactive oxygen species generation, and neutrophil extracellular trap (NET) formation15. Additionally, neutrophils are implicated in pulmonary microthrombosis, leading to vascular occlusion, impaired gas exchange, and hypoxic alveolar injury, further exacerbating lung damage through interactions with other immune cells16,17. Given that respiratory failure is a leading cause of mortality in bronchopulmonary cancer, our finding that neutrophil percentage is positively correlated with respiratory failure incidence highlights its role as a significant risk factor. Urea nitrogen (BUN) reflects alterations in microcirculation, renal dysfunction, and increased protein catabolism18,19. Malnutrition—common in bronchopulmonary cancer due to inadequate intake and high nutrient consumption by tumor cells—is associated with respiratory impairment20, suggesting that elevated BUN may indicate a heightened risk of respiratory failure. Hyperglycemia has been linked to poorer survival in multiple cancers21–23. A large prospective study (2011) reported consistent associations between fasting glucose > 100 mg/dL and worse outcomes in non-vascular malignancies24, aligning with our results. The underlying mechanisms warrant further investigation. Although PTT and serum sodium also emerged as high-risk factors in our model, few previous studies have reported their association with respiratory failure in this population. These findings require validation through larger prospective trials. Race was identified as another important predictor. African Americans and Native Hawaiians experience higher incidence and mortality from bronchopulmonary cancer than Whites25, and Black and Hispanic patients have been shown to have elevated malignancy risk and reduced survival26. Structural determinants—including higher exposure to air pollution, occupational hazards, and unequal access to healthcare—likely contribute to these disparities27,28. Consistent with these reports, our model indicates that African American and other non-White patients are at increased risk of respiratory failure compared to White patients. COPD contributes to respiratory failure through impaired diffusion capacity, inspiratory muscle fatigue, and alveolar hypoventilation29. Shock predisposes patients to respiratory failure by causing hypoxemia secondary to inadequate cardiac output and tissue perfusion30. Additionally, respiratory rate—a well-established discriminator of disease severity31,32—showed a stronger association with respiratory failure than heart rate in some studies33. Our results confirm that COPD, shock, elevated respiratory rate, and elevated heart rate are significant predictors of respiratory failure in bronchopulmonary cancer patients. However, the study has several limitations. First, because the majority of the population in the MIMIC database is Caucasian, there may be racial bias in the study population. Therefore, the applicability of our Nomograms to regions other than the United States remains to be investigated. Second, our data come from foreign databases, and the logistic regression prediction model is only internally validated, which may have geographical differences. Third, due to missing variables in the database, not all indicators related to the impact on respiratory function in bronchopulmonary cancer patients were included. The final population we selected for the study was patients between 2008 and 2019. The development of medical treatments and the optimization of treatment protocols during this period did not guarantee consistency in the approach to patient care, which may have affected the results of the study. Therefore, further prospective studies are needed to ensure the performance of our Nomograms. Conclusion Based on the information from the MIMIC database, we developed a nomogram to predict acute respiratory failure in bronchopulmonary cancer patients who were admitted to the ICU. Our validation analysis shows that the nomogram can provide clinicians with more information about the risk aspects of each prognostic factor. Supplementary Information Supplementary Information. Author contributions Ziwen Lv wrote the main manuscript text,Qi Feng prepared figures, Junhui Gu prepared table, Zhuo Gao,and Dong Liu prepared censor. Funding This work was supported by grants from the National Military Standards Program (BKJ20B047), Air Force Medical Center Science and Technology Boosting Program (2022ZTYB46), and Air Force Medical Center Clinic Study Program (2021LC006). Data availability The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request. Declarations Competing interests The authors declare no competing interests. Informed consent and ethical approval statement The studies involving human participants were reviewed and approved by the Institutional Review Board of the Massachusetts Institute of Technology and Beth Israel Deaconess Medical Center. Written informed consent for participation was not required for this study in accordance with the national legislation and institutional requirements. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Citations Supplementary Materials Supplementary Information. Data Availability Statement The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.