Evaluation of the PROSTia Test in Patients Undergoing Prostate Biopsy

.1. Outline of the study The lack of accuracy of the PSA test for prostate cancer screening leads to negative biopsies. The aim of this study is to determine whether the PROSTia test, a personalised medicine test using artificial intelligence (AI) by combining PSA, digital rectal examination (DRE) and 60 other qualitative data, could reduce the number of unnecessary biopsies and to estimate its impact on the detection of clinically significant cancers.

PROSTia employs the Gradient Boosting technique to select the variables of interest and optimise model performance. This process entails the utilisation of successive decision trees to model the non-linear relationships between the input variables and cancer risk. The construction of each tree is predicated on the correction of errors identified in preceding trees, with a focus on cases that have been misclassified. This iterative process is instrumental in generating a robust and accurate model for the purpose of prostate cancer screening

2 Methodology

The result of the PROSTia test is a score on a scale of 0 to 2. A score greater than or equal to 1 is considered a positive result. A positive result means that the patient has a significant risk of developing prostate cancer in the next 12 years. The following statistical analyses are performed using a 95% confidence interval:

• Sensitivity
• Specificity
• Positive likelihood ratio
• Negative likelihood ratio
• Disease prevalence
• Positive predictive value
• Negative predictive value
• Accuracy

Sensitivity, specificity, disease prevalence, positive and negative predictive values and accuracy are expressed as percentages.

Confidence intervals for sensitivity, specificity and accuracy are Clopper-Pearson 'exact' confidence intervals.

Confidence intervals for likelihood ratios are calculated using the logarithmic method as described on page 109 of Altman et al. 2000. Confidence intervals for predictive values are standard logit confidence intervals according to Mercaldo et al. 2007, except when the predictive value is 0 or 100%, in which case a Clopper-Pearson confidence interval is reported.

1. Number of patients The investigators set the number of patients at 150. This is a synthetic comparative study of a single patient cohort (i.e. there is no intervention as such; the cohort is compared to what it would be if the PROSTia test had been available).

Here are the assumptions that led to this number. Primary endpoint: binary (presence vs absence of disease).

Expected proportions: 50% in the study population (p2 without PROSTia) and 70% with PROSTia (p1).

Significance level (α): set at 0.05 Desired power (1-β): set at 0.8

Effect size: (p1-p2) drop rate:

The investigators plan to increase the calculated sample size to account for dropouts. calculated to account for dropouts or non-compliance with certain questionnaires (5%).

The method used to calculate the sample size is that of Woodward (1992).

1. Inclusion criteria

• Patients followed up at the Nancy CHRU in the Urology Department and eligible for prostate biopsy to screen for prostate cancer, taking into account an increase in PSA, PSA density,
• Adult males over 18 years of age
• Patients affiliated to or benefiting from a social security scheme
• Patients who understand French and are able to fill in a self-administered questionnaire or have someone to help them do so.

1. Non-inclusion criteria

• Patient who has already undergone prostate biopsy
• Patient under court protection, guardianship or trusteeship
• Patient deprived of liberty by judicial or administrative decision

1. Estimated time: Participation per person: Time taken to complete the questionnaire on the day of the biopsy of the biopsy (< 30 min) Duration of the inclusion period 12 months Total study duration 18 months