Table 4.

Estimated and observed proportions of clinically significant prostate cancer

		Patients without pathological confirmation	Patients with pathological confirmation
PI-RADS	Facility	Estimation	Observation	Estimation bias (estimation - observation in out of bag)
1–2	I	15.3% [12.2–18.5%]	17.6% (87/494)	−0.6% [−6.8 to 5.9%]
	II	12.6% [8.8–16.7%]	13.4% (29/216)	0.5% [−9.0 to 10.0%]
	III	13.5% [10.3–16.8%]	14.8% (49/331)	0.0% [−8.0 to 7.7%]
3	I	31.4% [27.7–34.9%]	34.3% (214/623)	−0.2% [−7.2 to 7.0%]
	II	18.4% [13.5–23.5%]	15.4% (24/156)	1.2% [−9.6 to 11.8%]
	III	28.6% [22.3–35.1%]	29.9% (46/154)	0.4% [−14.3 to 14.8%]
4	I	64.6% [61.8–67.2%]	65.3% (685/1049)	−0.0% [−5.9 to 5.5%]
	II	45.7% [41.8–49.8%]	47.5% (215/453)	0.6% [−8.2 to 9.9%]
	III	69.9% [66.2–73.6%]	67.5% (371/550)	−0.2% [−8.7 to 8.0%]
5	I	88.3% [86.4–90.2%]	87.8% (756/861)	−0.4% [−4.7 to 3.9%]
	II	75.4% [71.3–79.4%]	73.2% (267/365)	0.8% [−9.0 to 10.4%]
	III	86.1% [83.1–88.8%]	82.1% (343/418)	0.2% [−7.2 to 7.9%]

The estimated proportions of clinically significant prostate cancer (csPCa) are based on patients without pathological confirmation. The average of the model’s outputs is shown for each PI-RADS score, assuming a 100% pathological confirmation rate

The observed csPCa proportions are based on patients with pathological confirmation

The estimation biases represent the differences between the estimated and the observed csPCa proportions in patients with pathological confirmation

The 95% confidence intervals are shown in square brackets

PI-RADS Prostate Imaging-Reporting and Data System