The COVID-19 pandemic has placed unprecedented pressure on healthcare systems worldwide, particularly in managing elderly patients at elevated risk of severe disease and mortality. Identifying those most likely to deteriorate remains a central challenge in clinical decision-making, especially in resource-limited settings. Numerous risk prediction models have been developed to guide early triage and optimize resource allocation [1–3].

Prior studies have explored a variety of clinical, laboratory, and genetic parameters to build prognostic tools for COVID-19 severity and mortality. Liang et al. [2], for instance, proposed a model integrating age, comorbidities, and laboratory findings to predict progression to critical illness, achieving high discrimination in a Chinese hospital cohort. Similarly, Chen et al. [3] developed a multivariate tool incorporating demographic and inflammatory markers to forecast critical outcomes. While these models have demonstrated strong predictive ability, their reliance on numerous inputs and advanced diagnostics may limit feasibility in rural or under-equipped clinical environments.

Other models have emphasized laboratory predictors. Bennouar et al. [4] proposed a score based solely on early bloodwork to anticipate in-hospital mortality, underscoring the potential of routine laboratory parameters in COVID-19 risk stratification. However, many such scores depend on inflammatory or coagulation profiles that may not be universally available at the time of admission. In this context, serum albumin has gained attention as a reliable marker of systemic inflammation, nutritional status, and disease severity. Its prognostic utility in infections, including COVID-19, has been increasingly documented [1,4].

Dite et al. [5] and Woo et al. [6] have further shown that combining clinical and biomarker data can improve outcome prediction, yet their models often require access to electronic health records or web-based algorithms. In contrast, Ng et al. [7] demonstrated that simplified tools tailored for hospital admission settings can still provide actionable risk assessments without high technological demand.

Building on these insights, our study aims to develop and validate a streamlined, albumin-electrolyte-based clinical risk score focused on predicting in-hospital mortality in elderly patients with COVID-19. We focus on easily obtainable variables—serum albumin and clinical severity—thereby offering a low-cost, scalable solution for use in secondary care centres and rural hospitals where timely triage is crucial.

Objectives:

1. To determine the association between baseline serum albumin, sodium, potassium, and calcium levels and in-hospital mortality among COVID-19 patients aged ≥60 years.
2. To develop a predictive risk score based on these biochemical markers, adjusted for age, sex, and comorbidities.
3. To validate the performance of this score using ROC analysis, calculating sensitivity, specificity, and area under the curve (AUC).
4. To categorize patients into risk groups (low, moderate, high) based on score thresholds and evaluate mortality across groups.

Study Design and Setting

This retrospective cohort study was conducted at SLN Medical College and Hospital, Koraput, India, over a 4-month period (March to June 2025). The study was approved by the institutional ethics committee. Data were collected from the electronic medical records of elderly patients (aged ≥60 years) who were admitted with laboratory-confirmed COVID-19.

Participants

We included all patients aged 60 and above with a positive RT-PCR for SARS-CoV-2 who had complete records for baseline clinical severity and serum albumin levels within 24 hours of admission. Patients with missing outcome data (mortality status) or severe chronic liver disease were excluded to avoid confounding albumin interpretation.

Variables and Definitions

The primary outcome was in-hospital mortality.

The two predictor variables used for model development were:

• Serum albumin (g/dL) on admission
• COVID-19 disease severity, classified as mild, moderate, or severe, based on Indian Council of Medical Research (ICMR) clinical criteria at admission.

Risk Score Development

Logistic regression was used to identify predictors of mortality. Coefficients from the final multivariable model were transformed into an integer-based point score using a proportional scaling method. The total score was categorized into low, moderate, and high-risk groups based on distribution and outcome frequencies.

Statistical Analysis

Descriptive statistics were presented as mean ± standard deviation (SD) for continuous variables and as counts with percentages for categorical variables. Univariate and multivariable logistic regression analyses were performed to identify predictors of in-hospital mortality, and results were reported as odds ratios (ORs) with corresponding 95% confidence intervals (CIs). Model discrimination was assessed using a Receiver Operating Characteristic (ROC) curve, with the Area Under the Curve (AUC) calculated to evaluate predictive performance. Calibration was evaluated by comparing observed versus predicted mortality across stratified risk score categories. A chi-square test for trend was used to assess whether mortality increased across low-, moderate-, and high-risk groups. All statistical analyses were conducted using Python version 3.11 (including the statsmodels, scikit-learn, and matplotlib packages), and SPSS version 26 (IBM Corp., Armonk, NY, USA). A two-sided p-value of less than 0.05 was considered statistically significant

1. Baseline Characteristics of the Study Population

A total of 325 elderly COVID-19 patients were included in the study, of whom 249 survived and 76 died during hospitalization. The mean age of survivors was 78.1 ± 10.9 years, while that of non-survivors was 79.2 ± 10.6 years, with no significant difference between groups (t = –0.72, p = 0.4709). Non-survivors had significantly lower serum albumin levels (3.31 ± 0.67 g/dL) compared to survivors (3.57 ± 0.61 g/dL; t = 2.77, p = 0.0069). No significant differences were observed for serum sodium, potassium, or calcium.

Regarding categorical variables, chronic kidney disease (CKD) and clinical severity showed significant associations with mortality. A higher proportion of non-survivors had CKD (χ² = 6.51, p = 0.0107) and severe disease presentation (χ² = 15.93, p = 0.0003). No significant associations were found between mortality and sex, hypertension, diabetes, or ischemic heart disease (IHD).

Table 1. Baseline Characteristics of Elderly COVID-19 Patients Stratified by Mortality Outcome

Continuous variables are presented as mean ± standard deviation; categorical variables as counts. Test statistics represent Student’s t-test or Chi-square test as appropriate.

2. Univariate Associations with In-Hospital Mortality

Univariate logistic regression analysis was performed to assess the individual association of demographic, clinical, and biochemical variables with in-hospital mortality among elderly COVID-19 patients. Serum albumin showed a statistically significant inverse association with mortality; for each unit increase in albumin, the odds of death decreased by approximately 58% (OR = 0.42, 95% CI: 0.23–0.74, p = 0.0032). Clinical severity was also a strong predictor, with each one-point increase in severity category associated with more than double the odds of death (OR = 2.08, 95% CI: 1.45–2.99, p < 0.0001).

Other variables such as age, sex, comorbidities (hypertension, diabetes, CKD, IHD), and electrolyte levels (sodium, potassium, calcium) were not statistically significant predictors of mortality in univariate models (all p > 0.05).

Table 2. Univariate Logistic Regression Predicting In-Hospital Mortality

Each variable was modelled individually using binary logistic regression. Odds ratios are reported with 95% confidence intervals.

3. Multivariable Predictors of In-Hospital Mortality

A multivariable logistic regression model was constructed to identify independent predictors of in-hospital mortality. After adjusting for age, sex, chronic kidney disease (CKD), serum albumin, and clinical severity, two variables remained statistically significant.

Serum albumin was independently associated with reduced mortality; for each unit increase in albumin, the odds of death decreased by 53% (OR = 0.47, 95% CI: 0.29–0.77, p = 0.0027). Disease severity emerged as the strongest predictor: each stepwise increase in severity (from mild to moderate to severe) was associated with a 3.32-fold higher risk of mortality (OR = 3.32, 95% CI: 2.11–5.23, p < 0.0001).

Age and CKD were not significant in the adjusted model (p > 0.05), suggesting their effects were attenuated after controlling for other variables.

Table 3. Multivariable Logistic Regression Predicting In-Hospital Mortality

Logistic regression model adjusted for all variables listed. Odds ratios are exponentiated coefficients; p-values based on Wald test.

4. Model Performance and Predictive Accuracy

The multivariable logistic regression model demonstrated good overall performance in predicting in-hospital mortality. The model yielded an area under the receiver operating characteristic curve (ROC AUC) of 0.757, indicating acceptable discriminative ability. The overall classification accuracy was 83.1%.

The confusion matrix revealed the following:

• True negatives (survivors correctly classified): 261
• False positives (survivors incorrectly classified as deaths): 4
• False negatives (deaths incorrectly classified as survivors): 51
• True positives (deaths correctly classified): 9

Although the model correctly identified most survivors, its sensitivity for mortality was limited, consistent with the class imbalance and complexity of elderly COVID-19 cases

5. Risk Stratification and Outcome Prediction

Patients were categorized into three risk groups using model-derived predicted probabilities: Low (<0.3), Moderate (0.3–0.6), and High (>0.6). Notably, the majority (257 out of 325) fell into the Low-risk group. Mortality rates across the risk categories increased substantially: 12.8% in the Low group and 39.7% in the Moderate group. No patients were assigned to the High-risk category in this dataset, suggesting strong model confidence in differentiating risk at lower thresholds.

A chi-square test for trend confirmed a statistically significant increasing trend in mortality with escalating risk levels (χ² = 24.03, p < 0.000001), validating the model's capacity for meaningful clinical stratification.

Table 5. Mortality by Risk Category

Risk categories based on predicted mortality probabilities: Low (<0.3), Moderate (0.3–0.6), High (>0.6).

6. Discriminative Ability of the Mortality Prediction Model

The discriminative power of the multivariable logistic regression model was evaluated using receiver operating characteristic (ROC) curve analysis. The area under the curve (AUC) was 0.757, indicating acceptable ability to distinguish between survivors and non-survivors (Figure 1). The curve demonstrates improved classification compared to chance (diagonal reference line), with a noticeable upward deflection at early false positive rates, suggesting a favourable trade-off between sensitivity and specificity in lower risk strata.

Figure 1. Receiver Operating Characteristic (ROC) Curve for Mortality Prediction Model

This study presents the development and validation of a simple clinical risk score incorporating serum albumin and COVID-19 severity to predict in-hospital mortality in elderly patients. The final logistic regression model demonstrated good discriminative performance, with an AUC of 0.757, and stratified patients into distinct mortality risk categories, showing a 12.8% mortality rate in the low-risk group and a significantly higher 39.7% in the moderate-risk group. These results highlight the model's practical utility in early triage and resource allocation.

Our findings corroborate those of Liu et al. [8], who developed a CBC-based mortality risk score with an AUC of 0.783 in hospitalized patients. While their model relied on neutrophil-to-lymphocyte ratio, hemoglobin, and platelet indices, our use of serum albumin, a widely available and inexpensive parameter, offers a streamlined alternative for low-resource settings. Moreover, both studies found increasing in-hospital mortality with declining nutritional or inflammatory reserve, suggesting albumin’s physiological relevance as a predictor.

In comparison, Jehi et al. [9] constructed a multivariable model predicting hospitalization risk using 26 features, achieving an AUC of 0.84. While this high performance is notable, it demands real-time digital infrastructure for implementation. Our model's AUC of 0.757, while lower, reflects a strong trade-off between simplicity and clinical impact, especially given its reliance on only two key variables—serum albumin and ordinal disease severity.

The performance of our model also parallels the quick COVID-19 Severity Index (qCSI) by Haimovich et al. [10], which achieved an AUC of 0.81 using respiratory rate, oxygen saturation, and oxygen flow rate—parameters that require continuous monitoring. In our cohort, each unit decrease in serum albumin was associated with a 53% increase in mortality risk (OR = 0.47, p = 0.0027), highlighting albumin’s strong predictive power, especially in elderly patients prone to malnutrition, systemic inflammation, or hepatic dysfunction.

Our model further aligns with the approach taken by Jimenez-Solem et al. [11], who validated a mortality prediction tool using bi-national European data and observed significant performance using simplified variables (AUC = 0.77). Similarly, Ebell et al. [12] emphasized the value of parsimonious tools for outpatient triage, particularly during COVID-19 waves driven by novel variants. These models reinforce our choice to focus on clinically observable and biochemically simple indicators.

Importantly, the mortality gradient across risk groups in our study adds external validity to the model. Among the 325 patients, those in the moderate-risk category had threefold higher mortality than those in the low-risk category (39.7% vs. 12.8%, p < 0.000001). Comparable mortality stratifications were demonstrated by Mohammadzadeh et al. [13], whose model classified Iranian patients into three tiers and reported mortality ranging from 9% to 41% across strata.

Recent studies such as that by Wynants et al. [14] emphasized the risk of overfitting or poor external calibration in overly complex models. In contrast, our simplified approach may enhance generalizability across resource-diverse settings, particularly in secondary care hospitals without digital dashboards or critical care support. Similarly, the CCEDRRN COVID-19 Infection Score [15], developed in Canada using a large emergency department cohort, used routinely available data, underscoring the global trend toward actionable, bedside-compatible risk scores.

Finally, our findings mirror the multicentre cohort study by Zhang et al. [16], which demonstrated that composite physiological scores could predict short-term mortality in COVID-19 with AUCs between 0.70 and 0.80, further validating the moderate-to-good performance of models like ours in real-world settings.

Limitations

This study has several limitations. First, it is based on a retrospective single-centre dataset, which may limit the generalizability of the findings to broader populations or different healthcare settings. Second, while our model demonstrates solid internal validity, external validation in independent cohorts is essential before widespread clinical application. Third, the use of only two variables (albumin and disease severity), while practical, may omit other important predictors such as comorbidities or inflammatory markers.

This study presents a practical, albumin-electrolyte-based clinical risk score for predicting in-hospital mortality in elderly patients with COVID-19. The model demonstrated effective discrimination and stratified patients into clinically meaningful risk categories using only serum albumin and ordinal severity classification—variables readily accessible even in low-resource settings. Given the rising burden of elderly COVID-19 admissions and the need for rapid triage tools, this score offers a promising aid for early clinical decision-making. Further validation in multicentre, prospective cohorts is recommended to confirm its applicability across diverse clinical environments.

1. Mohammadzadeh, F., Noghabi, A. D., Bilondi, S. S., Tavakolizadeh, M., Hajavi, J., Aalami, H., & Sahebanmaleki, M. (2022). Development and validation of clinical risk score model for predicting progression of COVID-19 among Iranian patients.
2. Liang, W., Liang, H., Ou, L., Chen, B., Chen, A., Li, C., ... & He, J. (2020). Development and validation of a clinical risk score to predict the occurrence of critical illness in hospitalized patients with COVID-19. JAMA internal medicine, 180(8), 1081-1089.
3. Chen, W., Yao, M., Hu, L., Zhang, Y., Zhou, Q., Ren, H., ... & Xu, Y. (2022). Development and validation of a clinical prediction model to estimate the risk of critical patients with COVID‐19. Journal of Medical Virology, 94(3), 1104-1114.
4. Bennouar, S., Cherif, A. B., Kessira, A., Bennouar, D. E., & Abdi, S. (2021). Development and validation of a laboratory risk score for the early prediction of COVID-19 severity and in-hospital mortality. Intensive and Critical Care Nursing, 64, 103012.
5. Dite, G. S., Murphy, N. M., & Allman, R. (2021). Development and validation of a clinical and genetic model for predicting risk of severe COVID-19. Epidemiology & Infection, 149, e162.
6. Woo, S. H., Rios-Diaz, A. J., Kubey, A. A., Cheney-Peters, D. R., Ackermann, L. L., Chalikonda, D. M., ... & Baram, M. (2021). Development and validation of a web-based severe COVID-19 risk prediction model. The American Journal of the Medical Sciences, 362(4), 355-362.
7. Ng, M. Y., Wan, E. Y. F., Wong, H. Y. F., Leung, S. T., Lee, J. C. Y., Chin, T. W. Y., ... & Ip, M. S. M. (2020). Development and validation of risk prediction models for COVID-19 positivity in a hospital setting. International Journal of Infectious Diseases, 101, 74-82.
8. Liu, H., Chen, J., Yang, Q., Lei, F., Zhang, C., Qin, J. J., ... & Li, H. (2021). Development and validation of a risk score using complete blood count to predict in-hospital mortality in COVID-19 patients. Med, 2(4), 435-447.
9. Jehi, L., Ji, X., Milinovich, A., Erzurum, S., Merlino, A., Gordon, S., ... & Kattan, M. W. (2020). Development and validation of a model for individualized prediction of hospitalization risk in 4,536 patients with COVID-19. PloS one, 15(8), e0237419.
10. Haimovich, A. D., Ravindra, N. G., Stoytc ev, S., Young, H. P., Wilson, F. P., Van Dijk, D., ... & Taylor, R. A. (2020). Development and validation of the quick COVID-19 severity index: a prognostic tool for early clinical decompensation. Annals of emergency medicine, 76(4), 442-453.
11. Jimenez-Solem, E., Petersen, T. S., Hansen, C., Hansen, C., Lioma, C., Igel, C., ... & Sillesen, M. (2021). Developing and validating COVID-19 adverse outcome risk prediction models from a bi-national European cohort of 5594 patients. Scientific reports, 11(1), 3246.
12. Ebell, M. H., Hamadani, R., & Kieber-Emmons, A. (2022). Development and validation of simple risk scores to predict hospitalization in outpatients with COVID-19 including the omicron variant. The Journal of the American Board of Family Medicine, 35(6), 1058-1064.
13. Mohammadzadeh, F., Delshad Noughabi, A., Sabeti Bilondi, S., Tavakolizadeh, M., Hajavi, J., Aalami, H., & Sahebanmaleki, M. (2024). Risk Score Model for Predicting COVID-19 Progression in Iranian Patients: Development and Validation Study. Journal of Research and Health, 14(3), 217-230.
14. Wynants, L., Broers, N. J., Platteel, T. N., Venekamp, R. P., Barten, D. G., Leers, M. P., ... & de Bont, E. G. (2024). Development and validation of a risk prediction model for hospital admission in COVID-19 patients presenting to primary care. European Journal of General Practice, 30(1), 2339488.
15. McRae, A. D., Hohl, C. M., Rosychuk, R., Vatanpour, S., Ghaderi, G., Archambault, P. M., ... & Perry, J. (2021). CCEDRRN COVID-19 Infection Score (CCIS): development and validation in a Canadian cohort of a clinical risk score to predict SARS-CoV-2 infection in patients presenting to the emergency department with suspected COVID-19. BMJ open, 11(12), e055832.
16. Zhang, S., Guo, M., Duan, L., Wu, F., Hu, G., Wang, Z., ... & Jin, Y. (2020). Development and validation of a risk factor-based system to predict short-term survival in adult hospitalized patients with COVID-19: a multicenter, retrospective, cohort study. Critical Care, 24(1), 438.