A mathematical approach improves the predictability of length of hospitalization due to acute odontogenic infection: A retrospective investigation of 303 patients.

PURPOSE

Increasing rates of hospitalization of patients diagnosed with acute odontogenic infection have become a burden for public health care, with significant economic concerns. The aim of this study was to investigate factors that tend to prolong hospital length of stay (LOS) in the treatment of severe infections. We present a statistical model that enables the prediction of LOS by exposing the feasibility of the essential statistical determinants.

MATERIALS AND METHODS

A 5-year retrospective study investigated records of 303 in-hospital patients with abscess of odontogenic origin. Time-to-event models were used to analyse data where the outcome variable is the time to the occurrence of a specific event. Here, the focus is on a statistical model for the prediction of LOS of patients.

RESULTS

The group of all patients (n = 303) was analysed by considering seven characteristics of the patients (age, gender, spreading of infection, localization of infection focus, type of administered antibiotics, diagnosed diabetes mellitus, and existence of a remaining infection focus). Age (p = 0.049; rc = -0.007) and spreading of infection (p < 0.001; rc = -0.965) showed a significant impact on the LOS. Subjects were divided into two groups. Group A (n = 185) consisted of patients who presented with a severe odontogenic infection and not yet removed infection focus; group B were patients having undergone outpatient operative tooth removal (n = 118). To group A patients' data, two new risk factors ("days between abscess incision and removal of infection focus" = dbir and "removal of infection focus during the same stay as abscess incision" = riss) replaced the risk factors "remaining infection focus." A significant impact on the LOS was detected for dbir (p < 0.001; rc = -0.15) and riss (p < 0.001; rc = -1.76). Our statistical model explicitly describes how the probability for discharge depends on the time and how specific characteristics affect the LOS. We observed a significantly higher LOS in older patients and subjects with infection spreading. In group A patients, dbir and riss had a highly significant impact on the LOS.

CONCLUSION

Predicting the LOS may promote transparency to costs and management of patients under inpatient treatment. Our statistical model describes the probability of a discharge at time t compared to a discharge later than t (a LOS longer than t). Furthermore, the model enables a prediction of the LOS of each patient for practitioners in an easy way.