AKI and Delirium Risk Prediction and Cost Burden Study

This study is a retrospective observational study designed to develop and validate early risk prediction models for two common hospital-acquired complications: acute kidney injury (AKI) and delirium. These complications frequently occur during hospitalization and are associated with prolonged hospital length of stay (LOS), increased intensive care unit (ICU) utilization, higher health care costs, and worse clinical outcomes. Early identification of patients at elevated risk may enable targeted preventive strategies and support future clinical decision support interventions.

The study will use routinely collected data from the electronic health record (EHR) system at Nebraska Medicine. The study has two primary objectives. The first primary objective is to develop and temporally validate prediction models that estimate the risk of hospital-acquired AKI and delirium that is accurate, well calibrated, clinically usable, and fair among important population subgroups, using information available early during hospitalization. The second primary objective is to quantify the hospital Length of Stay and direct cost burden attributable to these complications.

Data Sources and Predictor Variables:

Structured EHR data will be extracted for eligible inpatient encounters. Candidate predictors will include demographic characteristics, comorbid conditions, baseline kidney function, early laboratory values, vital signs, nursing and therapy flowsheets, medication exposures (including nephrotoxic and psychoactive medications), admission service, and indicators of physiologic instability recorded during the early hospitalization period.

Predictor variables will be defined using information available within the first 0-24 hours of admission to ensure temporal ordering between predictors and subsequent complications. Standardized clinical definitions and coding systems will be used where applicable.

Outcome Identification:

Hospital-acquired AKI will be identified using Kidney Disease Improving Global Outcomes (KDIGO) creatinine-based criteria derived from laboratory data in the EHR. Delirium will be identified using validated clinical screening results including Confusion Assessment Method (CAM) or Brief Confusion Assessment (bCAM) or CAM-ICU assessments when available, supplemented by occupational therapy and nursing flowsheets, and diagnostic codes documented in the medical record. Only complications occurring after the defined early hospitalization window will be included to preserve temporal validity.

Prediction Model Development:

Prediction models for hospital-acquired AKI and delirium will be developed using both regression-based and machine learning approaches.

Statistical methods will include:

* Regularized logistic regression (LASSO) to perform variable selection and reduce overfitting when evaluating multiple candidate predictors.
* Extreme Gradient Boosting (XGBoost) to capture nonlinear relationships and complex interactions between predictors.
* Additional model comparison approaches may include random forest or elastic net regression where appropriate.

Model performance will be evaluated using multiple complementary metrics:

* Discrimination: area under the receiver operating characteristic curve (AUROC) and precision-recall area under the curve.
* Calibration: calibration slope, calibration-in-the-large, and graphical calibration plots.
* Overall model performance: Brier score.
* Clinical utility: decision curve analysis evaluating net benefit across clinically relevant risk thresholds.
* Fairness: equitable performance across major population subgroups with subgroup analysis

Internal validation will be conducted using bootstrap resampling and k-fold cross-validation to estimate optimism-corrected model performance and assess potential overfitting.

Analysis of Hospital Utilization and Cost Outcomes:

The study will also estimate the incremental hospital utilization and cost burden associated with incident hospital-acquired AKI and hospital-acquired delirium. It will also examine related outcomes, including transfer to an intensive care unit, hospital mortality, and 30-day readmission.

Because AKI and delirium may develop at different times during hospitalization, each condition will be analyzed as an event that can change a patient's hospital course. Follow-up will begin at hospital admission. Patients with the condition already present at admission will be excluded from the analysis of that condition.

These complications will be treated as time-varying exposures. To preserve temporal ordering between predictors, complications, and downstream outcomes, analyses will use a multistate analysis approach to represent movement through clinically relevant stages of hospitalization, such as admission without the complication, development of hospital-acquired AKI or delirium, transfer to an intensive care unit, discharge, or death. At each hospital day, patients who develop the complication will be compared with similar patients who are still hospitalized and have not yet developed it. This approach accounts for the timing of the complication and reduces bias that could occur if patients were classified only according to whether they experienced the complication at any time during hospitalization.

Comparisons will be made with patients who remain hospitalized and complication-free at the same time point using risk-set comparison methods. The analysis will estimate the difference in hospital days and costs between the observed course of care and an estimated comparison scenario in which the hospital-acquired complication did not occur. Hospital costs will be evaluated using daily cost information and statistical methods appropriate for healthcare cost data.

Outcome models will include:

* Hospital length of stay: negative binomial regression modeling residual LOS following complication onset.
* ICU transfer and 30-day readmission: multivariable logistic regression.
* Hospital costs: two-part modeling approach consisting of logistic regression for the probability of incurring costs and generalized linear models with gamma distribution and log link for positive cost values.

To reduce confounding, analyses will adjust for baseline covariates measured prior to complication onset. Differences in patient characteristics and clinical condition before the complication occurs will be addressed using multivariable adjustment and inverse probability weighting to standardize comparisons between exposed and unexposed patients and estimate incremental attributable outcomes. These methods will improve comparability between patients who develop hospital-acquired AKI or delirium and patients who remain free of the corresponding complication. Sensitivity analysis and subgroup analyses will be performed to assess the consistency and robustness of the findings.

Sensitivity and Subgroup Analysis:

Sensitivity analyses will evaluate robustness of findings using alternative modeling strategies for time-varying exposures, including marginal structural models with stabilized inverse probability weights. Additional analyses will assess potential model performance differences across clinically relevant subgroups such as age, baseline kidney disease status, and clinical service lines.

Data Quality and Validation:

Data quality procedures will include structured data extraction protocols, range checks, and consistency checks across related data fields. Variable definitions will follow standardized clinical definitions where possible, and analytic datasets will be validated against source EHR data to ensure completeness and accuracy.

Study Purpose:

This study uses existing clinical data and does not alter patient care. The findings will provide foundational evidence for the development of EHR-integrated risk prediction tools and clinical decision support strategies aimed at improving early identification and prevention of hospital-acquired complications such as AKI and delirium.

• It is observational trial with no intervention being evaluated (Other): Hospital-acquired acute kidney injury and hospital-acquired delirium will be identified retrospectively from electronic health record data using prespecified clinical definitions. Acute kidney injury will be defined using KDIGO-based criteria, and delirium will be identified using a prespecified phenotype incorporating available CAM-family assessments and other time-stamped structured clinical documentation and flowsheets. Patients with the relevant condition present on admission will be excluded from that outcome analysis. Neither exposure is assigned by the investigators; each will be analyzed separately as a time-varying complication according to its documented onset.

Primary outcomes

• Incremental Hospital Days Attributable to Hospital-Acquired Acute Kidney Injury [From inpatient admission to live discharge, in-hospital death, or day 30, whichever occurs first; assessed up to 30 days.]: Difference in restricted mean hospital days per hospitalization between the observed course and a standardized hypothetical scenario in which hospital-acquired AKI did not occur, estimated using the prespecified multistate analysis. Reported in days, with a 95% confidence interval.
• Incremental Direct Hospital Cost Attributable to Hospital-Acquired Acute Kidney Injury [From inpatient admission to live discharge, in-hospital death, or day 30, whichever occurs first; assessed up to 30 days.]: Difference in cumulative direct hospital cost per hospitalization between the observed course and the standardized hypothetical AKI-prevention scenario. Reported in 2025 US dollars, with a 95% confidence interval.
• Incremental Hospital Days Attributable to Hospital-Acquired Delirium [From inpatient admission to live discharge, in-hospital death, or day 30, whichever occurs first; assessed up to 30 days.]: Difference in restricted mean hospital days per hospitalization between the observed course and a standardized hypothetical scenario in which hospital-acquired delirium did not occur, estimated using the prespecified multistate analysis. Reported in days, with a 95% confidence interval.
• Incremental Direct Hospital Cost Attributable to Hospital-Acquired Delirium [From inpatient admission to live discharge, in-hospital death, or day 30, whichever occurs first; assessed up to 30 days.]: Difference in cumulative direct hospital cost per hospitalization between the observed course and the standardized hypothetical delirium-prevention scenario. Reported in 2025 US dollars, with a 95% confidence interval.
• Temporal Validation Performance of Hospital-Acquired Acute Kidney Injury Prediction Models [Prediction at hour 24 for incident HA-AKI during hours 24-96 after admission]: Locked HA-AKI prediction models will be evaluated in the 2024-2025 temporal-validation cohort. Using information available by hour 24, model will predict HA-AKI during hours 24-96. Performance will be assessed using AUROC for discrimination and calibration slope for agreement between predicted and observed risks. Estimates will be reported with 95% confidence intervals.
• Temporal Validation Performance of Hospital-Acquired Delirium Prediction Model [Prediction at hour 24 for incident hospital-acquired delirium during hours 24-96 after admission]: Locked hospital-acquired delirium prediction model will be evaluated in the 2024-2025 temporal-validation cohort. Using information available by hour 24, model will predict complication during hours 24-96. Performance will be assessed using AUROC for discrimination and calibration slope for agreement between predicted and observed risks. Estimates will be reported with 95% confidence intervals.

Secondary outcomes

• Incidence of Hospital-Acquired Acute Kidney Injury [From hour 24 after inpatient admission until live discharge, in-hospital death, or hospital day 30, whichever comes first, assessed up to 30 days]: Number and percentage of eligible hospitalizations in which incident HA-AKI develops after the prespecified present-on-admission period. HA-AKI will be identified using the locked KDIGO-based definition and assigned to the earliest qualifying onset time. Results will be reported overall and by AKI stage.
• Incidence of Hospital-Acquired Delirium [From hour 24 after inpatient admission until live discharge, in-hospital death, or hospital day 30, whichever comes first, assessed up to 30 days.]: Number and percentage of eligible hospitalizations in which incident hospital-acquired delirium develops after the prespecified present-on-admission period. Delirium will be identified using the locked electronic health record phenotype and assigned to the earliest qualifying onset time.
• Cumulative Incidence of Intensive Care Unit (ICU) Transfer Following Hospital-Acquired Acute Kidney Injury (HA-AKI) [From HA-AKI onset until ICU transfer, live discharge, in-hospital death, or hospital day 30, whichever comes first (assessed up to 30 days after admission)]: Percentage of participants with HA-AKI who are transferred to an ICU. This is estimated as the cumulative incidence by day 30 after HA-AKI onset, using a competing-risks method (Aalen-Johansen estimator) with in-hospital death as a competing event and live discharge as censoring. The analysis includes only participants who were not already receiving intensive care at HA-AKI onset. Unit of Measure: Percentage of participants
• Cumulative Incidence of Intensive Care Unit (ICU) Transfer Following Hospital-Acquired Delirium [From hospital-acquired delirium onset until ICU transfer, live discharge, in-hospital death, or hospital day 30, whichever comes first (assessed up to 30 days after admission)]: Percentage of participants with hospital-acquired delirium who are transferred to an ICU. This is estimated as the cumulative incidence by day 30 after delirium onset, using a competing-risks method (Aalen-Johansen estimator) with in-hospital death as a competing event and live discharge as censoring. The analysis includes only participants who were not already receiving intensive care at delirium onset. Unit of Measure: Percentage of participants
• In-Hospital Mortality Associated with Hospital-Acquired Acute Kidney Injury and Delirium separately [From inpatient admission until in-hospital death, live discharge, or hospital day 30, whichever comes first, assessed up to 30 days.]: In-hospital death will be evaluated separately for HA-AKI and hospital-acquired delirium. Death will be modeled as a terminal event distinct from live discharge. Standardized mortality probabilities will be reported under observed care and the corresponding hypothetical complication-prevention scenario, with 95% confidence intervals.
• Live Discharge Following Hospital-Acquired Acute Kidney Injury and Delirium separately [From inpatient admission until live discharge, in-hospital death, or hospital day 30, whichever comes first, assessed up to 30 days]: The probability and timing of live hospital discharge will be evaluated separately for HA-AKI and hospital-acquired delirium. Live discharge and in-hospital death will be treated as distinct terminal outcomes in the multistate analysis. Standardized live-discharge probabilities will be reported with 95% confidence intervals.
• All-Cause 30-Day Hospital Readmission [Within 30 days after live discharge from the index hospitalization]: Number and percentage of patients discharged alive who experience an all-cause inpatient readmission within 30 days after discharge from the index hospitalization. Readmission will be evaluated separately in relation to HA-AKI and hospital-acquired delirium. Post-discharge death will be handled as a competing event if sufficiently complete vital-status data are available.
• Additional Temporal Validation Performance of the Prediction Models [Prediction at hour 24 for incident HA-AKI or hospital-acquired delirium during hours 24-96 after admission]: Additional performance of the locked HA-AKI and delirium prediction models will be evaluated separately using area under the precision-recall curve, calibration-in-the-large, observed-to-expected ratio, flexible calibration curves, Brier score, and scaled Brier score. Outcome prevalence and predicted-risk distributions will also be reported, with patient-level 95% confidence intervals where applicable.
• Component-Specific Hospital Costs Attributable to Hospital-Acquired Acute Kidney Injury and Delirium [From inpatient admission until live discharge, in-hospital death, or hospital day 30, whichever comes first, assessed up to 30 days.]: Incremental costs within reliable, mutually exclusive hospital cost categories will be estimated separately for HA-AKI and hospital-acquired delirium. Categories may include room and board, intensive care, pharmacy, laboratory testing, imaging, procedures, consultations, nursing or patient-care services, and renal replacement therapy. Results will be reported in 2025 US dollars with 95% confidence intervals.
• Clinical Utility of the Hospital-Acquired Acute Kidney Injury and Delirium Prediction Models [Prediction at hour 24 for incident HA-AKI or hospital-acquired delirium during hours 24-96 after admission]: Clinical utility will be evaluated separately for each model across prespecified risk thresholds using decision-curve net benefit. At selected thresholds, sensitivity, specificity, predictive values, likelihood ratios, false-positive and false-negative rates, alerts per 100 admissions, true-positive and false-positive alerts per 100 admissions, and number needed to evaluate will be reported.
• Fairness and Subgroup Performance of the Prediction Models [Prediction at hour 24 for incident HA-AKI or hospital-acquired delirium during hours 24-96 after admission]: Prediction performance will be evaluated within prespecified patient groups defined by age, sex, race, ethnicity, preferred language, insurance, service, initial care location, calendar period, CKD or eGFR for the AKI model, and cognitive impairment or dementia for the delirium model. Representation, missingness, prevalence, calibration, discrimination, prediction error, threshold-specific errors, alert burden, and net benefit will be reported with confidence intervals.
• Severity of Hospital-Acquired Acute Kidney Injury and Delirium [From complication onset until live discharge, in-hospital death, or hospital day 30, whichever comes first, assessed up to 30 days after admission.]: HA-AKI severity will be described using the highest KDIGO stage reached during the index hospitalization. Delirium severity or motor subtype will be described when supported by sufficiently complete and validated electronic health record data. These analyses will characterize complication heterogeneity and will not be interpreted as baseline causal subgroup effects.

Other outcomes

• Internal Validation Performance of the Prediction Models [Prediction at hour 24 for incident HA-AKI or hospital-acquired delirium during hours 24-96 after admission]: Internal validation of the HA-AKI and delirium prediction models will use patient-level bootstrap resampling of the 2019-2023 development cohort. Apparent and optimism-corrected AUROC, area under the precision-recall curve, calibration intercept and slope, Brier score, and scaled Brier score will be reported. The complete model-development process will be repeated within resampling when computationally feasible.
• Supportive Encounter-Level Length-of-Stay and Cost Associations [From inpatient admission (total length of stay) or complication onset (post-onset length of stay) until live discharge, in-hospital death, or hospital day 30, whichever comes first, assessed up to 30 days]: Supportive adjusted regression analyses will estimate associations between HA-AKI or hospital-acquired delirium and total or post-onset hospital length of stay and encounter-level direct hospital cost. These analyses will complement the primary time-resolved multistate analyses but will not be interpreted as fully time-resolved attributable effects.
• Prediction Performance at Alternative Prediction Origins and Horizons [Predictions at hours 12 or 48 for outcomes during the prespecified subsequent period through day 7 or discharge]: As sensitivity analyses, prediction performance may be assessed using alternative prediction origins, including hours 12 and 48, and an alternative outcome horizon through day 7 or discharge. These analyses represent separate prediction use cases and will not be described as validation of the primary hour-24 models.
• Attributable Burden Within Prespecified Patient Subgroups [From inpatient admission until live discharge, in-hospital death, or hospital day 30, whichever comes first, assessed up to 30 days.]: Incremental restricted mean hospital days and, when estimable, direct hospital costs will be reported within prespecified baseline subgroups. Subgroups include age, sex, race, ethnicity, service, initial care location, calendar period, baseline severity, CKD or eGFR for HA-AKI, and cognitive impairment or dementia for delirium. Exposure-by-subgroup interactions and standardized contrasts will be reported with confidence intervals.

Sex: All, ages: 19 Years to any

Inclusion Criteria:

1. Age 19 years or older at the time of admission.
2. Inpatient admission to the general medical (non-ICU) ward at Nebraska Medical Center, Omaha, Nebraska.
3. Admission date between January 1, 2019 to December 31, 2025.

Exclusion Criteria:

1 . Admitted directly to Intensive Care unit. 2. Received Mechanical Ventilation at the time of admission. 3. Dependent on maintenance Dialysis at the time of admission (excluded from acute kidney injury analysis only).

4\. Acute Kidney Injury present on admission (excluded from acute kidney injury analysis only) .

5\. Delirium present on admission (excluded from Delirium analysis only) .

• Silver SA, Chertow GM. The Economic Consequences of Acute Kidney Injury. Nephron. 2017;137(4):297-301. doi: 10.1159/000475607. Epub 2017 Jun 9. PMID 28595193
• Koyner JL, Carey KA, Edelson DP, Churpek MM. The Development of a Machine Learning Inpatient Acute Kidney Injury Prediction Model. Crit Care Med. 2018 Jul;46(7):1070-1077. doi: 10.1097/CCM.0000000000003123. PMID 29596073
• van den Boogaard M, Pickkers P, Slooter AJ, Kuiper MA, Spronk PE, van der Voort PH, van der Hoeven JG, Donders R, van Achterberg T, Schoonhoven L. Development and validation of PRE-DELIRIC (PREdiction of DELIRium in ICu patients) delirium prediction model for intensive care patients: observational multicentre study. BMJ. 2012 Feb 9;344:e420. doi: 10.1136/bmj.e420. PMID 22323509
• Hshieh TT, Yue J, Oh E, Puelle M, Dowal S, Travison T, Inouye SK. Effectiveness of multicomponent nonpharmacological delirium interventions: a meta-analysis. JAMA Intern Med. 2015 Apr;175(4):512-20. doi: 10.1001/jamainternmed.2014.7779. PMID 25643002
• Rubin FH, Neal K, Fenlon K, Hassan S, Inouye SK. Sustainability and scalability of the hospital elder life program at a community hospital. J Am Geriatr Soc. 2011 Feb;59(2):359-65. doi: 10.1111/j.1532-5415.2010.03243.x. PMID 21314654
• Leslie DL, Marcantonio ER, Zhang Y, Leo-Summers L, Inouye SK. One-year health care costs associated with delirium in the elderly population. Arch Intern Med. 2008 Jan 14;168(1):27-32. doi: 10.1001/archinternmed.2007.4. PMID 18195192
• Zeng X, McMahon GM, Brunelli SM, Bates DW, Waikar SS. Incidence, outcomes, and comparisons across definitions of AKI in hospitalized individuals. Clin J Am Soc Nephrol. 2014 Jan;9(1):12-20. doi: 10.2215/CJN.02730313. Epub 2013 Oct 31. PMID 24178971
• Silver SA, Long J, Zheng Y, Chertow GM. Cost of Acute Kidney Injury in Hospitalized Patients. J Hosp Med. 2017 Feb;12(2):70-76. doi: 10.12788/jhm.2683. PMID 28182800
• Marcantonio ER. Delirium in Hospitalized Older Adults. N Engl J Med. 2017 Oct 12;377(15):1456-1466. doi: 10.1056/NEJMcp1605501. PMID 29020579
• Witlox J, Eurelings LS, de Jonghe JF, Kalisvaart KJ, Eikelenboom P, van Gool WA. Delirium in elderly patients and the risk of postdischarge mortality, institutionalization, and dementia: a meta-analysis. JAMA. 2010 Jul 28;304(4):443-51. doi: 10.1001/jama.2010.1013. PMID 20664045
• Coca SG, Yusuf B, Shlipak MG, Garg AX, Parikh CR. Long-term risk of mortality and other adverse outcomes after acute kidney injury: a systematic review and meta-analysis. Am J Kidney Dis. 2009 Jun;53(6):961-73. doi: 10.1053/j.ajkd.2008.11.034. Epub 2009 Apr 5. PMID 19346042
• Chertow GM, Burdick E, Honour M, Bonventre JV, Bates DW. Acute kidney injury, mortality, length of stay, and costs in hospitalized patients. J Am Soc Nephrol. 2005 Nov;16(11):3365-70. doi: 10.1681/ASN.2004090740. Epub 2005 Sep 21. PMID 16177006
• Inouye SK, Westendorp RG, Saczynski JS. Delirium in elderly people. Lancet. 2014 Mar 8;383(9920):911-22. doi: 10.1016/S0140-6736(13)60688-1. Epub 2013 Aug 28. PMID 23992774
• Susantitaphong P, Cruz DN, Cerda J, Abulfaraj M, Alqahtani F, Koulouridis I, Jaber BL; Acute Kidney Injury Advisory Group of the American Society of Nephrology. World incidence of AKI: a meta-analysis. Clin J Am Soc Nephrol. 2013 Sep;8(9):1482-93. doi: 10.2215/CJN.00710113. Epub 2013 Jun 6. PMID 23744003
• Hoste EAJ, Kellum JA, Selby NM, Zarbock A, Palevsky PM, Bagshaw SM, Goldstein SL, Cerda J, Chawla LS. Global epidemiology and outcomes of acute kidney injury. Nat Rev Nephrol. 2018 Oct;14(10):607-625. doi: 10.1038/s41581-018-0052-0. PMID 30135570