Adriele da Mota de França Pinto; Wagner Wilson Bortoletto

DOI: 10.22167/2675-6528-202602601

Article derived from a Course Conclusion Work (TCC), with content based on the student’s original work and adapted to the editorial format of the E&S Magazine with the support of the ResumeAI tool, an artificial intelligence solution developed by Instituto Pecege for textual synthesis and organization.

Summary

The Intensive Care Unit (ICU) represents one of the most expensive hospital sectors in the healthcare system. Regression models can assist managers in decision-making by identifying factors that impact healthcare costs. The objective was to analyze the association between direct hospital cost ranges and sociodemographic, clinical, and input variables in patients admitted to an ICU in a hospital in Paraná. A cross-sectional and retrospective study was conducted, using secondary data from hospital audits between September 2023 and September 2024. Inferential analyses were applied, with Chi-Square and Kruskal-Wallis tests, and computational analysis, developing a multinomial logistic regression model. It was observed that patients with a clinical profile, use of enteral nutrition, transfusion, radiography, tomography, and death were associated with Medium and High cost ranges. Age, use of echocardiogram, and the admission ICD of Chapter 6 – Diseases of the nervous system were associated only with the High cost range. It was concluded that the sociodemographic, clinical, and input factors used by ICU patients, associated with cost ranges, can support decision-making by hospital managers and guide the development of clinical protocols.

Keywords: Regression Analysis; Direct Costs of Services; Intensive Care Units.

1. Introduction

Hospital costs represent a significant component of the healthcare system, being defined as all expenses necessary for the provision of services. This category encompasses both direct costs, which are objectively measurable from patient expense statements, and indirect costs, which require approximations and allocation criteria as they are not individually attributable (Ministry of Health [MS], 2013).

In Brazil, the health sector absorbs approximately 9.5% of the Gross Domestic Product, with 1.9% of this total directed to hospital expenses. Hospitals are recognized as complex and costly institutions (Almeida et al., 2024). Within this scenario, the Intensive Care Unit (ICU) stands out as one of the most expensive sectors, responsible for critically ill patients. Although ICUs represent only ten percent of hospital beds, they can account for a share of eight to thirty percent of a hospital’s total expenses (Bruyneel et al., 2023).

A multicenter study conducted in Brazilian ICUs revealed that the total average cost per patient, encompassing both direct and indirect costs, reached approximately R$15,906.51 (Bass et al., 2025). However, these costs can vary considerably depending on several factors, such as the occurrence of hospital-acquired infections, patient age, length of stay, presence of comorbidities, and procedures performed (Reddy et al., 2024; Bass et al., 2025). For example, patients who develop healthcare-associated infections may incur a high average cost of R$58,053.13 (Bass et al., 2025).

The wide variability of hospital costs, influenced by clinical variables and patient prognostic evolution, highlights the importance of developing predictive models. Such models are fundamental to identify the variables that impact direct costs in critical patients, aiding in resource management and more efficient decision-making (Gomes and Borgert, 2025).

In developed countries, such as France, the United States, and Belgium, the analysis of healthcare-related costs is frequently performed using regression models, whether linear or logistic (Vimont et al., 2022). Academic literature has widely documented the use of regression models to investigate variables associated with high hospital costs, due to their interpretative capacity and applicability in clinical practice (Ruijter et al., 2024; Marín et al., 2025).

Regression models, compared to other approaches, allow for a clearer estimation of the impact of independent variables on the dependent variable. They are capable of directly quantifying how much a given factor increases or reduces the probability of a specific outcome (Ruijter et al., 2024; Marín et al., 2025). This ability to identify and quantify factors is crucial for assisting hospital management in optimizing and allocating resources (Marín et al., 2025).

The development of regression models to analyze the association between costs incurred in ICUs and clinical variables is, therefore, a valuable tool. It allows for the identification of factors influencing costs, supports decision-making, optimizes resource allocation, formulates specific strategies for hospital management, and potentially predicts future hospital costs (Gomes and Borgert, 2025; Langenberger et al., 2023). Given the above, the objective of the present work is to analyze the association between hospital direct cost ranges and sociodemographic, clinical, and input-use-related variables in patients admitted to an Intensive Care Unit of a hospital in the state of Paraná.

2. Material and Methods

The present study was characterized as quantitative, observational, cross-sectional, and retrospective, with the objective of analyzing the association between direct hospital cost ranges and sociodemographic, clinical, and input variables. The research was conducted with secondary data from hospitalizations that occurred in an Intensive Care Unit (ICU) of a university hospital in the state of Paraná, Brazil.

Data collection occurred between September 2023 and September 2024, using information from a database provided by the institution’s Hospital Audit department. Sociodemographic (age and sex), clinical (initial diagnosis, care profile, infection development, and length of stay), and utilized supplies (exams, procedures, and hospital materials) data were collected. Direct costs attributed to patients were obtained from the contractual value of the supplies multiplied by the quantity consumed during hospitalization.

The initial diagnosis was categorized according to the International Classification of Diseases – 10th Revision (ICD-10), assigned by the on-call physicians upon admission (Wells et al., 2017), with low-frequency chapters grouped into “Other chapters”. The length of stay considered every ICU admission as one day, following the hospital’s internal regulations. The patient profile, classified as clinical or surgical, was determined by the procedure groups of Ordinance GM/MS No. 2,848/2007 (Brazil, 2007).

Hospital exams, procedures, and materials were extracted from the Tasy® system. Healthcare-Associated Infections (HAIs) were classified according to the criteria of the National Health Surveillance Agency (ANVISA, 2017), including Ventilator-Associated Pneumonia (VAP), Central Line-Associated Bloodstream Infection (CLABSI), and Catheter-Associated Urinary Tract Infection (CAUTI). Clinical evolution was categorized as sectoral discharge or death.

For inclusion in the sample, patients aged 18 years or older with complete information for all variables were considered. Data with duplicate attendances and direct cost equal to R$0.00 were excluded. Of 562 initial patients, 484 individuals were included for convenience.

Direct costs were categorized as Low cost, Medium cost, and High cost, based on the distribution quartiles. Low cost comprised values less than or equal to the first quartile (Q1), Medium cost included values greater than Q1 and less than the third quartile (Q3), and High cost encompassed values greater than or equal to Q3 (Marín et al., 2025). Sociodemographic, clinical, and input variables were defined as independent, and direct cost ranges as dependent.

The data analysis was carried out in two stages: statistical and computational. In the statistical stage, descriptive analysis was conducted (absolute and relative frequencies for categorical variables; measures of central tendency and dispersion for numerical variables). For inferential analysis, the association between categorical variables was evaluated by the Chi-squared test (Fávero and Belfiore, 2025). For numerical variables, the data distribution was tested by the Kolmogorov-Smirnov test, which indicated non-normality. Thus, for the total direct cost ranges, the Kruskall-Wallis and Dunn post hoc tests were applied (Fávero and Belfiore, 2025). The significance level adopted for all analyses was 5%.

In the computational stage, a Multinomial Logistic Regression model was developed, suitable for a qualitative response variable with more than two categories (Fávero and Belfiore, 2025). This supervised machine learning model estimated the probability of each observation belonging to a category of the outcome variable, defining one category as a reference for comparison of the others via odds ratio [OR], evaluating the individual effect of each explanatory variable (Fávero and Belfiore, 2025).

The evaluation of the final model used Log-Likelihood (LL), chi-square statistic (X²), and McFadden’s Pseudo-R², which indicated the model’s fit and likelihood gain compared to a null model (Fávero and Belfiore, 2025). The overall model performance was assessed by accuracy, sensitivity, specificity metrics, and the area under the curve (AUC) of the predicted classifications, for both the general model and “One-vs-All” (OVA) comparisons (Abramovich et al., 2020).

3. Results and Discussion

This study sought to analyze the association between direct hospital cost ranges and various sociodemographic, clinical, and input variables in patients admitted to an Intensive Care Unit. The results obtained, through statistical and computational analyses, reveal important patterns that can support hospital management in resource optimization and strategic decision-making. The final sample analyzed comprised 484 individuals, whose data were categorized and subjected to inferential tests and multinomial logistic regression modeling.

Statistical analysis

Descriptive analyses of the categorical data revealed that the majority of patients were male, representing 53.93% of the sample, and presented a clinical profile in 56.00% of cases. The most prevalent admission diagnosis was Chapter 19 – Injuries and external causes, corresponding to 31.20% of patients. Regarding the use of supplies, high utilization of catheters (76.24%), enteral diet (65.91%), radiography (70.25%), and tomography (56.61%) was observed. The predominant clinical outcome was discharge from the hospital unit, with 80.37% of patients, while 19.6% evolved to death. These findings are in line with retrospective studies in ICUs with clinical and surgical profiles, which also report a prevalence of male patients and traumatic diagnoses (Krieger et al., 2024; Michetti et al., 2019).

Regarding the numerical data, the median age of the patients was 63 years, with a standard deviation of 20 years, and the median ICU length of stay was 4 days, with a standard deviation of 7 days. These values are compatible with the literature, which frequently describes ICU patients with a mean age around 58 years and a median length of stay of 5 days, in addition to significant use of hospital supplies and a lower prevalence of Healthcare-Associated Infections (HAIs) (Krieger et al., 2024; Michetti et al., 2019).

Inferential analyses of direct cost bands, using Chi-Square and Kruskall-Wallis tests, indicated significant associations for most variables, except sex (p=0.330). Patients in the Low direct cost band were predominantly those with diagnoses from Chapter 13 – Musculoskeletal diseases (63.3%) and Other chapters (100%), and who did not use enteral nutrition (49.7%), radiography (49.3%), or tomography (49.0%). In contrast, the High direct cost band was associated with patients who received transfusions (56.5%), ultrasonography (46.4%), electroencephalogram (57.5%), and who developed IPCS-AC (100.0%), PAV (87.1%), and ITU-AC (57.1%). The remaining variables showed a higher prevalence in the Medium direct cost band.

Age analysis did not reveal a statistically significant difference between direct cost ranges (p=0.122). However, ICU length of stay showed a highly significant difference (p<0.001). Patients with Low cost had a median stay of 1 day, while those with Medium cost stayed for 4 days and those with High cost for 14 days. These results reinforce the literature that points to the influence of sociodemographic, clinical, and care factors on hospital costs in ICUs, with higher costs associated with invasive devices, prolonged length of stay, neurological and cardiovascular pathologies, and infections (Bruyneel et al., 2025; Mastrogianni et al., 2021).

Computational analysis

For the computational analysis, a multinomial logistic regression model was developed, with the Low cost category as the reference. The final model, after the application of the stepwise procedure, proved to be statistically significant (χ²=403.82; p<0.001). The Log-Likelihood was -301.31, representing a substantial improvement compared to the null model (-503.22), and the Pseudo-R² of 0.40 indicated a reasonable fit of the model to the observed data (Fávero and Belfiore, 2025). This performance suggests that the set of included predictors significantly improves the ability to explain the variation in cost ranges.

The final model’s performance metrics revealed an overall accuracy of 71.9%, sensitivity of 69.8%, and specificity of 83.8%. The ROC curves showed Area Under the Curve (AUC) values of 0.93 for Low cost, 0.81 for Medium cost, and 0.89 for High cost. These AUC values are considered good to excellent, especially above 0.80, indicating robust model performance in predicting high-cost patients (Langenberger et al., 2023; Dilixiati et al., 2024). High specificity is particularly valuable for hospital management, as it allows for the correct identification of patients who do not belong to the high-cost category, avoiding unnecessary interventions and optimizing resource allocation (Dilixiati et al., 2024; Kaiser et al., 2021).

The analysis of the multinomial logistic regression model coefficients identified several factors associated with cost ranges. Patients with clinical profile (OR=3.14; 95%CI=1.69; 5.82), who used enteral diet (OR=5.64; 95%CI=2.98; 10.67), received transfusion (OR=9.32; 95%CI=2.92; 29.71), underwent radiography (OR=2.92; 95%CI=1.58; 5.40) and tomography (OR=9.72; 95%CI=4.87; 19.39), and who progressed to death (OR=5.65; 95%CI=1.76; 18.12) showed a higher probability of belonging to the Medium direct hospital cost range (p<0.01 for all variables). These results underscore the complexity of the factors driving moderate costs in intensive care settings.

For the High direct cost group, the probability was significantly higher for patients with clinical profile (OR=4.07; 95%CI=1.78; 9.33), who used enteral diet (OR=50.89; 95%CI=15.74; 164.49), received transfusion (OR=49.18; 95%CI=13.79; 175.31), underwent radiography (OR=26.41; 95%CI=7.49; 93.07) and tomography (OR=18.88; 95%CI=7.58; 47.03), and who died (OR=12.19; 95%CI=3.25; 45.75). Furthermore, age (OR=0.97; 95%CI=0.95; 0.99), the use of echocardiogram (OR=4.38; 95%CI=1.30; 14.76), and the admission ICD belonging to Chapter 6 – Diseases of the nervous system (OR=11.49; 95%CI=1.02; 129.34) were also associated with High direct cost (p<0.05 for all variables).

The association of age with a lower probability of High direct cost (OR=0.97) is a notable finding and consistent with studies indicating that older patients, especially those over 80 years old, tend to be less likely to generate high costs (Dziegeilewski et al., 2021). This can be explained by the greater fragility and multimorbidity in the elderly, which often result in shorter hospital stays in acute care and fewer interventions, thus reducing accumulated costs (Zhang et al., 2025; Dziegeilewski et al., 2021).

Patients with a clinical profile showed a higher chance of belonging to the Medium and High cost categories compared to surgical patients. This result corroborates the literature, which suggests that clinical patients generally have more comorbidities, greater disease severity, and longer hospital stays, consequently demanding more health interventions (Bruyneel et al., 2025; Gershengorn et al., 2015). The complexity of clinical cases requires a more intensive allocation of resources, directly impacting hospital costs.

The admission diagnosis related to Chapter 6 – Diseases of the nervous system (ICD) was associated with an 11.49 times higher probability of belonging to the High direct cost category. Neurological diseases frequently require complex hospitalizations, with multiple laboratory tests, surgical interventions and invasive procedures, in addition to extensive imaging exams (Raj et al., 2024). This finding suggests the need to review institutional clinical protocols, focusing on strategies such as early rehabilitation and the use of bedside ultrasound, which can contribute to cost reduction (Shrestha et al., 2025; Aung et al., 2019).

The use of enteral diet was associated with higher chances of belonging to the Medium and High cost categories. This result contrasts with national studies in the United States, where early enteral nutrition was associated with reduced total ICU costs, lower mortality, and shorter duration of mechanical ventilation and hospital stay (Dainelli et al., 2023; Haines et al., 2023). The divergence may indicate an inadequate hospital protocol in the analyzed institution, such as late initiation of nutrition, use of expensive formulas without proportional benefit, or lack of specific immunonutrition options, aspects that hospital management can optimize (Haines et al., 2023; Dainelli et al., 2023).

Patients who received blood transfusions also showed increased chances of belonging to the Medium and High cost categories. A Brazilian study demonstrated that transfused individuals had 5.1 times higher costs, and a Swiss study associated the transfusion of a single unit of packed red blood cells with a significant increase in hospital costs (Machado et al., 2019; Saporito et al., 2026). It is recommended that hospital managers guide care teams to adopt restrictive transfusion strategies, based on criteria established by the Ministry of Health, to reduce unnecessary transfusions and prevent complications (Saporito et al., 2026; Machado et al., 2019).

The performance of radiography and tomography was associated with higher chances of belonging to the Medium and High cost categories. Although these exams contribute to costs, a study in hospitals in Belgium indicated that diagnostic imaging represented an insignificant fraction of total daily costs compared to length of stay and the use of mechanical ventilation (Bruyneel et al., 2023). Thus, the review of routine imaging practices, based on evidence, can reduce average monthly hospital costs by up to 37% without impacting mortality or ICU length of stay (Scott et al., 2020; Lee et al., 2015).

The use of echocardiogram was also associated with higher chances of belonging to the High cost category. A survey of more than 390 U.S. hospitals revealed that institutions with higher rates of echocardiography utilization had average costs approximately US$ 3,164 higher per hospitalization (Pack et al., 2019). The adoption of a more judicious and selective use of echocardiogram, avoiding prolonged hospitalization solely for the examination without expectation of a change in management, can generate economic gains without compromising clinical outcomes (Pack et al., 2019).

Finally, patients whose clinical evolution resulted in death showed increased chances of belonging to both the Medium cost and High cost categories. This finding is aligned with the literature, which shows that patients who die may generate costs up to 12.4% higher than survivors (Bruyneel et al., 2023; Kramer et al., 2017). This is due to the intensive use of clinical interventions until the exhaustion of therapeutic possibilities, including high-cost medications, prolonged mechanical ventilation, invasive procedures, and management of multiple infections, which prolong hospitalization time (Bruyneel et al., 2023; Kramer et al., 2017).

Although hospital services should provide the resources to reverse pathological conditions, it is crucial that the care team is trained to recognize the prognosis and the possibility of disease irreversibility early on (Quinn et al., 2021). A care approach focused on end-of-life quality of life can contribute to the rational use of hospital resources and to the provision of care more suited to the patient’s clinical needs and therapeutic goals (Quinn et al., 2021). The developed model, therefore, offers a tool to improve care protocols, focus on cost-effective interventions, and develop predictive tools for financial risk at hospital admission (Marín et al., 2025).

In summary, the results obtained demonstrate that sociodemographic, clinical, and input factors are crucial for determining direct costs in ICUs. The identification of these associations, especially through the multinomial logistic regression model, allows hospital managers a deeper understanding of cost drivers. This understanding is fundamental for the development of more efficient clinical protocols, the optimization of resource allocation, and strategic decision-making that reconciles economic efficiency with the quality of care and the ethical principles of distributive justice (Charlton, 2022; Bilinski et al., 2022), directly responding to the objective of analyzing the association between cost ranges and the studied variables.

4. Conclusion

This study aimed to analyze the association between direct hospital cost ranges and sociodemographic, clinical, and input variables in patients admitted to an Intensive Care Unit of a hospital in Paraná. It was found that a multinomial logistic regression model demonstrated robust capacity to identify factors that significantly impact care costs. It was observed that patients with a clinical profile, use of enteral feeding, blood transfusion, radiography and tomography, and death were associated with higher probabilities of belonging to the Medium and High direct cost ranges. Additionally, age, use of echocardiogram, and admission diagnosis related to nervous system diseases (Chapter 6 of ICD) were specifically associated with the High cost range. The overall accuracy of the model, at 71.9%, and Area Under the Curve (AUC) values above 0.80 for all cost categories, indicated consistent predictive performance, capable of supporting hospital management in the early identification of patients with higher cost potential.

Despite the relevance of the findings, the study, which is cross-sectional and retrospective in nature and conducted in a single institution, limits the generalization of results to other hospital contexts. However, the identification of predictor variables for cost offers a substantial practical contribution, by allowing hospital managers to review and improve institutional clinical protocols, especially regarding the use of enteral nutrition, transfusions, imaging tests and echocardiograms, and the management of patients with neurological diseases and at the end of life. It is suggested that future studies explore the implementation of predictive tools for financial risk at hospital admission, based on this model, and evaluate their impact on optimizing resource allocation and quality of care. The developed model therefore represents a valuable tool for strategic decision-making, aiming to reconcile economic efficiency with the quality of care in ICUs.

Bibliographic References

Almeida, J.F.F.; Conceição, S.V.; Magalhães, V.S.; Alemão, M.M. 2024. Estimating and planning hospital costs of public hospitals in Brazil. The International Journal of Health Planning and Management 39(6): 1790-1809.

Bass, L. M.; Henrique, L.; Kiriyama, E.J.; Oliveira, N.; Honorato, A.; Cristina, K.; Moraes, R.; Prandini, C.M.; Silva, C.; Gonsalez, R.; Franco, F. F.; Petenate, A.J.; Reis, M.; Garcia, C.; Vernal, S. 2

Bruyneel, A.; Larcin, L.; Martins, D.; Van, J.; Leclercq, P.; Pirson, M. 2023. Cost comparisons and factors related to cost per stay in intensive care units in Belgium. BMC Health Services Research 23(1).

Ministério da Saúde [MS]. 2013. Introdução à Gestão de Custos em Saúde. 1ed. Ministério da Saúde, Brasília, GO, Brasil. Disponível em: <https://www.gov.br/saude/pt-br/acesso-a-informacao/gestao-do-sus/economia-da-saude/nucleos-de-economia-da-saude/publicacoes>. Acesso em: 17 agosto 2025.

Article originating from the Final Course Work of the Specialization in Data Science and Analytics of the MBA USP/Esalq

To learn more about the course, click here and access the MBX Academy platform