Description

Hatipoglu et al reported a model for predicting all-cause readmission within 30 days for a patient admitted with pneumonia. This can help to identify a patient who may benefit from more aggressive management. The authors are from the Cleveland Clinic.


Patient selection: admitted to the hospital with pneumonia

 

Outcome: 30-day all-cause readmission

 

Parameters:

(1) age in years

(2) history of cancer

(3) history of coronary heart disease (CHD)

(4) history of stroke

(5) time until first antibiotic administered in minutes

(6) therapy with opioids

(7) maximum temperature in °C during 24 hours after admission

(8) maximum serum albumin in g/dL during the 24 hours after admission

(9) maximum serum BUN in mg/dL in the last 24 hours prior to discharge

(10) minimum serum sodium in mmol/L in the last 24 hours prior to discharge

(11) minimum hemoglobin in g/dL in the last 24 hours prior to discharge

(12) minimum INR in the last 24 hours prior to discharge

(13) total number of prior admissions within the 6 months prior to admission

 

Parameter

Finding

Points

age in years

 

0.0143 * (age)

history of cancer

no

0

 

yes

0.229

history of CHD

no

0

 

yes

0.221

history of stroke

no

0

 

yes

-0.192

time to first antibiotic

 

0.140 * LN(time)

therapy with opioids

no

0

 

yes

0.375

maximum temperature

 

-0.104 * (temperature)

maximum BUN

 

0.308 * (BUN)

minimum sodium

 

-0.0358 * (sodium)

minimum hemoglobin

 

-0.130 * (hemoglobin)

minimum albumin

 

-0.342 * (albumin)

minimum INR

 

0.32 * (INR)

number of admissions

 

0.689 * (number)

 

where:

The beta-coefficient for prior admissions is ambiguous. The coefficient associated with the odds ratio is 0.689. There is a second that was fitted with a restricted spline given as -1.03. There are 3 options - use the first, use the second, use both.

Serum BUN appears to be a key predictor for readmission.

 

X =

= SUM(points for all 13 parameters) - 12.3

 

probability of readmission =

= 1/ (1+EXP((-1) * X))

 

Performance:

The area under the ROC curve is 0.71.


To read more or access our algorithms and calculators, please log in or register.