A model to compute the medical cost of patients in intensive care

Objective: Our objective was to identify, among the information routinely collected on patients in intensive care units (ICUs), data that determine the total cost for a given patient. Design: We developed a model that could help physicians in medical ICUs to estimate the cost of care for their patients when no cost data were available at the individual patient level. Setting: A Medical ICU. Patients and participants: The model was developed using a random sample of 73 patients admitted to the medical ICU in 1996 and 1997, validated by another random sample of 29 patients admitted during the same period. Interventions: The actual medical variable cost per patient was computed from data on the total resources used (excluding personnel and fixed costs), collected from the patients' records plus pharmacy, laboratory and blood bank logs. The explanatory variables tested were: length of stay, nursing workload, severity of condition, and procedures recorded by a score [omega (Omega)] including 3 components related to the frequency of procedure use. The model was constructed in a stepwise fashion, assuming a linear relation. Equations were tested on the basis of the residual mean square; criteria for inclusion and elimination of variables were the level of its partial regression coefficient and medical criteria. The model was validated by analysis of variance of the regression on a second population of 29 patients using the F-test. Main outcome measures and results: The median length of stay was 7 days (range: 3 to 22 days). Mortality rate was 25%. Median medical variable cost was pound 805 (mean medical variable cost was pound 1738, total cost was pound 6279). The variables selected in the multiple regression model as relevant predictors of medical costs were: procedures recorded only once during the ICU stay irrespective of their reiteration (Omega(1)), procedures recorded every time they an performed (Omega(2)), procedures recorded daily in the ICU (Omega(3)) and the presence or absence of an invasive procedure (Kc). The final equation, calibrated with r(2) of 0.826 and p > 0.0001, was: medical cost (pound) = 23 Omega(1) + 53 Omega(2) + 8 Omega(3) + 2352Kc + 96. The validation with the other sample of 29 patients compared actual to predicted costs. Analysis of variance of the regression from the model was r(2) = 0.596 (p > 0.05). Conclusions: Our standardised cost model is a possible approach to allow comparison of medical costs within and between ICUs.