Decision curve analysis based on summary data.

Area of Science:

• Biostatistics
• Clinical Decision Making
• Precision Medicine

Background:

• Precision medicine requires integrating predictive models with decision analysis frameworks like decision curve analysis (DCA).
• Current DCA applications necessitate individual patient data (IPD), which is often inaccessible.
• Developing methods for aggregate data DCA can enhance precision medicine adoption.

Purpose of the Study:

• To present a statistical framework enabling DCA using only aggregate data (mean and standard deviation) from predictive model probabilities.
• To validate this aggregate-data approach by comparing it with traditional IPD-based DCA through simulations and real-world datasets.

Main Methods:

• Developed a statistical framework for aggregate data DCA using mean and standard deviation of predictive model probabilities.
• Conducted extensive simulations to compare aggregate-based DCA with IPD-based DCA.
• Applied the framework to four diverse predictive models using IPD from studies on statins, hospice referral, thromboprophylaxis, and sinusoidal obstruction syndrome prevention.

Main Results:

• Simulations showed negligible differences between aggregate and IPD-based DCA when predictive models were well-calibrated.
• For adequately powered models, aggregate data DCA closely approximated IPD-derived DCA results.
• A larger discrepancy was observed between aggregate and IPD-based DCA for models with smaller sample sizes due to inherent instability.

Conclusions:

• DCA using summary statistics (mean and SD) from adequately powered and calibrated models closely approximates IPD-based DCA.
• The use of aggregate data significantly broadens the applicability of DCA, overcoming IPD accessibility limitations.
• This approach promotes the integration of predictive and decision modeling for advancing individualized patient care.