Table 4 Studies developing risk models or clinical decision rules

León et al., 2006 [27]

Risk factors significant (P < 0.05) on univariable analysis of full sample included in multivariable logistic regression model fitted in 65% development sample

Final model chosen by backward elimination; stopping criterion unclear

Simplified to bedside score by rounding coefficients.

ROC and sensitivity/specificity at cut-off values in 35% validation sample

0.908 × (total parenteral nutrition) + 0.997 × (surgery) + 1.112 × (multifocal Candida species colonisation) + 2.038 × (severe sepsis)

1 × (total parenteral nutrition) + 1 × (surgery) + 1 × (multifocal Candida species colonisation) + 2 × (severe sepsis)

Bedside score ≥ 3, or, equivalently, severe sepsis plus at least one other risk factor or all three other risk factors

Ostrosky-Zeichner et al., 2007 [37]

All rule development in 75% development sample

Univariable analysis of risk factors

Clinical decision rules constructed for 'all possible combinations of risk factors and time points' in 'several different formats (with different weights for the risk factors)'

'Best' rules selected on sensitivity, PPV, PPV/(1-NPV), and proportion of patients identified as high risk in development sample

χ² test of association, sensitivity, specificity, PPV and NPV in 25% validation sample

Any antibiotic days 1 to 3 and CVC days 1 to 3

Any antibiotic days 1 to 3) and CVC days 1 to 3 and at least one of the following:

Any surgery days -7 to 0

Immunosuppressive use days -7 to 0

Pancreatitis days -7 to 0

TPN day 1-3; dialysis day 1-3

Steroid use days -7 to 3

Any antibiotic days 1 to 3 or CVC days 1 to 3 and at least two of the following:

Any surgery days -7 to 0

Immunosuppressive use days -7 to 0

Pancreatitis days -7 to 0

TPN days 1 to 3

Dialysis days 1 to 3

Steroid use days -7 to 3

Paphitou et al., 2005 [35]

All rule development in full sample

Univariable analysis of risk factors; unclear whether used to select factors for multivariable model

Multivariable logistic regression model with stepwise procedure; unclear whether forward selection or backward elimination and unclear stopping criterion

Clinical decision rules constructed 'using a combination of inspection of the data and results of the multivariable analysis'

Sensitivity and PPV in full sample

NNT assuming 50% relative reduction in IFD associated with treatment

Cost to prevent one case assuming prophylaxis $100/day

At least one of the following:

Diabetes mellitus

TPN days -7 to 0

New-onset haemodialysis days -7 to 3

At least one of the following:

Diabetes mellitus

TPN days -7 to 0

New-onset haemodialysis days -7 to 3

Broad-spectrum antibiotics days -7 to 3

At least one of the following:

Diabetes mellitus

TPN days -7 to 0

New-onset haemodialysis days -7 to 3

and

Broad-spectrum antibiotics days -7 to 3

Pittet et al., 1994 [33]

All rule development in full sample

Clinical decision rules constructed from colonisation parameters only (number of sites, colonisation index, corrected colonisation index derived post hoc); methods unclear

Risk factors with P < 0.15 on univariable analysis included in multivariable logistic regression model

Only those with P < 0.05 in multivariable model reported; unclear if stepwise procedure used

Clinical decision rules validated by sensitivity, specificity, PPV and NPV in full sample

Colonisation at two or more sites

Colonisation at three or more sites

Colonisation at four or more sites

Candida colonisation index ≥ 0.5

Candida corrected colonisation index ≥ 0.4