Discriminant Analysis of Rasmussen et al for Predicting Outcome in Acute Renal Failure

 

Variables used for stepwise discriminant analysis:

(1) pre-existing heart disease: includes coronary artery disease, valvular heart disease or congestive heart failure

(2) acute heart disease: includes myocardial infarction, myocarditis or infective endocarditis

(3) neoplastic disease: malignant solid or hematologic tumors not considered cured

(4) pre-existing renal disease: baseline serum creatinine concentration > 1.7 mg/dL prior to development of acute renal failure

(5) oliguria: urine output < 400 mL per 24 hours

(6) acute pancreatitis: serum amylase > 1,000 IU/L with appropriate clinical setting

(7) trauma, including severe burns

(8) "other" surgery: any procedure done using general anesthesia other than aortic aneurysm repair or open heart surgery.

(9) CNS damage

(10) respiratory failure

Predictor Variable	Status	Function Coefficient
pre-existing heart disease	absent	0
	present	0.1371
acute cardiac illness	absent	0
	present	0.4248
neoplastic disease	absent	0
	present	0.3957
pre-existing renal disease	absent	0
	present	0.2117
oliguria	absent	0
	present	0.3746
acute pancreatitis	absent	0
	present	0.3498
trauma	absent	0
	present	0.3170
"other" surgery	absent	0
	present	0.2086
CNS damage	absent	0
	present	0.1977
respiratory failure	absent	0
	present	0.1960

 

discriminant score =

= SUM(points for all 10 variables) – 0.5194

 

where:

• -0.5194 is a constant.

 

Interpretation:

• minimum score: -0.5194

• maximum score: 2.2216

• The cutoff for the discriminant score was 0.21, with a score > 0.21 indicating a fatal outcome.

 

Performance with a cutoff of 0.21:

• positive predictive value: 100%

• sensitivity: 58% (32 of 78 nonsurvivors had scores <= 0.21)

 

Limitations:

• The study was performed in 1985 and survival may have improved since then.

• When a test group was evaluated using the equation, a cutoff of 0.46 was needed to give a 100% positive predictive value, with a sensitivity of 26%.