Mitch Medical Healthcare

Laparoscopic Partial Nephrectomy Greater or Lesser Than One Day Duration of Hospital Stay Predictor

Univariate logistic regression analysis in the design group for duration of hospital stay greater or lesser than one day identified 24 patient characteristics as significant predictors for a duration of hospital stay of greater than one day as listed in Table 6. As there were only 28 patients in the design group who had a duration of hospital stay of less than or equal to one day, the statistical significance of some of these parameters may be overstated.

Multiple logistic regression analysis (Table 6) identified congeorine heart failure, parenchymal tumor, and tumor abutting the collecting system to be independent significant predictors of a duration of hospital stay greater than one day. This analysis identified the appropriate weighting for characteristics in the design of a linear regression prediction algorithm. The following parameters were utilized in the model: hypertension, gas-tic esophageal reflux disease, anxiety, hematuria (micro or gross), diabetes mellitus, CRI, gout, congestive heart failure, Crohn's disease, transabdominal approach, solitary kidney,

TABLE 6 ■ Laparoscopic Partial Nephrectomy Group: Analysis for Duration of Hospital Stay Less Than or Greater Than One Day

Predictor of less than or	Odds	Univariate		Odds	Multivariate
greater than 1 day	ratio	95% CI	Paval	ratio	95% CI	Paval
Hypertension	0	0	<0.001	0.8	0.2-4.1	0.793
History of deep venous thrombus	TO	TO	<0.001	23K	0—TO	0.997
Peripheral vascular disease	0	0	<0.001	0.01	0—TO	0.999
Gastro-esophageal reflux disease	0	0	<0.001	385K	0—TO	0.992
Anxiety	TO	TO	<0.001	90K	0—TO	0.996
Depression	TO	TO	<0.001	178K	0-TO	0.997
Liver cirrhosis	0	0	<0.001	0	0—TO	0.998
Hepatitis C	TO	TO	<0.001	0.04	0—TO	0.999
Hematuria	TO	TO	<0.001	8M	0—TO	0.998
COPD	0	0	<0.001	6M	0—TO	0.993
Alcohol use	TO	TO	<0.001	11K	0—TO	0.997
Hypothyroidism	TO	TO	<0.001	12M	0—TO	0.995
Diabetes	0	0	<0.001	10M	0—TO	0.991
Renal insufficiency	TO	TO	<0.001	24M	0—TO	0.996
Stroke/TIA	TO	TO	<0.001	0.08	0—TO	1.000
Gout	TO	TO	<0.001	37B	0—TO	0.999
Congestive heart failure	0	0	<0.001	0.002	0—0.06	<0.001
Osteoarthritis	TO	TO	<0.001	77K	0—TO	0.997
Thrombocytopenia	TO	TO	<0.001	2M	0—TO	0.996
Crohn's disease	0	0	<0.001	0.2	0—325	0.657
Obstructive sleep apnea	0	0	<0.001	0.4	0—89	0.755
Anemia	TO	TO	<0.001	382K	0—TO	0.999
Transabdominal approach	TO	TO	<0.001	2.9	0.8—10.3	0.093
Solitary kidney	TO	TO	<0.001	72M	0—TO	0.997
Exophytic tumor	0	0	<0.001	0.06	0.01—0.3	<0.001
Tumor extends up to renal sinus	TO	TO	<0.001	1.5	0.3—8.4	0.648
Tumor abuts collecting system	0	0	<0.001	0.02	0—0.1	<0.001

Note: Predictors in italics were used to create the model.

aP values in this table refer to the comparison of those patients that had a post-op duration of stay less than 1 day versus those greater than 1 day.

Abbreviations: COPD, chronic obstructive pulmonary disease; CI, confidence interval; TIA, transient ischemic attack.

Note: Predictors in italics were used to create the model.

aP values in this table refer to the comparison of those patients that had a post-op duration of stay less than 1 day versus those greater than 1 day.

Abbreviations: COPD, chronic obstructive pulmonary disease; CI, confidence interval; TIA, transient ischemic attack.

exophytic tumor, tumor extending up to the renal sinus and tumor abutting the collecting system. The combination of these factors provided the optimal accuracy. The addition of the other factors listed in Table 4 did not enhance model accuracy, and thus they were not utilized.

The equation for the lesser or greater than one day duration of hospital stay prediction model is:

1 Day score = Hypertension+Gastic erophageal reflux disease + Anxiety + Hematuria + Diabetes mellitus + CRI + Gout + Congestive heart failure+ Crohn's + Transabdominal approach + Solitary kidney — Exophytic tumor + Up to sinus + Abuts collecting system

Key:

Presence of any of the above conditions: Yes = 1, no = 0.

If the score is greater than 0.5, then a duration of hospital stay of greater than one day is predicted. If the score is lesser than 0.5, a duration of hospital stay of less than or equal to one day is predicted. The model accuracy in the design group was 83% (receiver operating characteristic 0.8) and in the test group was 84% (receiver operating characteristic 0.8) as shown in Table 7.

In order to simplify the use of these four models Palm™- and Windows™-based versions were created.

TABLE 7 ■ Accuracy Profiles for the Duration of Hospital Stay Prediction Models

	Greater or lesser		Greater or lesser
	than 2 day		than 1 day
DOS model	Accuracy (%)	ROC	Accuracy (%) ROC
Laparoscopic radical/simple nephrectomy
Design group (393 pts)	74	0.8	73 0.7
Test group (29)	66	0.7	97 0.8
Laparoscopic partial nephrectomy
Design group (344 pts)	73	0.7	83 0.8
Test group (19)	68	0.6	84 0.8
Abbreviations: ROC, receiver operating characteristic; DOS, duration of hospital stay.

The results illustrate that there are certain preoperative patient parameters that may predict a longer DOS in patients undergoing laparoscopic radical/simple or partial nephrectomy.

Our design was based on strict statistical methods of univariate and multivariate analysis and logistic regression.

Comment: The results illustrate that there are certain preoperative patient parameters that may predict a longer duration of hospital stay in patients undergoing laparo-scopic radical/simple or partial nephrectomy.

These parameters may help guide physicians when counseling patients considering these procedures. The models are a user-friendly tool to provide predictions as to the expected duration of hospital stay. These models may be implemented in clinical care pathways to provide efficient use of medical resources during the hospital stay. Two completely separate sets of models (using separate patient groups) were designed for the radical/simple nephrectomy patients and the partial nephrectomy patients, as these are two very different procedures.

The problem with such models is that they tend to be biased to the patient population at the design institution. Therefore, multi-institutional external validation and refinement of the models would enhance their utility. We are currently performing this phase of development.

There are some confounding variables that may be decreasing the predictive accuracy of the models: subjective, but important factors such as baseline patient coping capacity, motivational level of the patient and level of surgeon optimism. However, these factors are difficult to objectively quantify into modeling parameters. Despite these caveats, the models do provide fair predictions based on the patient parameters that were evaluated in this study.

Our design was based on strict statistical methods of univariate and multivariate analysis and logistic regression (24,25).

The accuracy of the models and the receiver operating characteristic values did vary in the test groups compared to the design groups. This may be in part due to the small sample size of the testing groups.

In order to facilitate free exchange and testing of the model, Hand-held and Windows-based computer versions of the model were created. A web-based distribution platform was designed and the programs can be downloaded as free shareware. (from www.uroengineering.com). This allows physicians to use the models in their practice and to validate the effectiveness of the models at their institution.

Conclusions: These models provide 66% to 97% accuracy in predicting the postla-paroscopic nephrectomy duration of hospital stay. These models may allow the urologist to preoperatively counsel patients and to optimize the delivery of care during the hospital stay.

Our group is currently in the process of performing a multi-institutional prospective testing, refinement and validation study for these models. This will assess and develop the model's true widespread clinical utility.

Mathematical modeling in laparoscopy and urology will only open more opportunities to further optimize the delivery of care to our patients.