Michael P. Dykstra

PURPOSE: Large language models (LLMs) may improve extraction of prognostic variables in prostate cancer from unstructured clinical text compared with traditional, rule-based natural language processing. METHODS: We used iterative prompt engineering with few-shot examples to develop LLM prompts for 30 phenotypes from prostate biopsy, radical prostatectomy (RP), and transurethral resection of the prostate (TURP) pathology reports, as well as magnetic resonance imaging (MRI) pelvis, computed tomography [CT] abdomen/pelvis, Tc-99m bone scan, and prostate-specific membrane antigen [PSMA] PET/CT reports. Data were drawn from >130 Veterans Affairs facilities (1999-2025). Inference was performed with Llama 3.3 70B or GPT-4o depending on the task. Performance was evaluated on independent test sets with metrics including overall accuracy, sensitivity, positive predictive value (PPV), negative predictive value (NPV), and macro-F1. RESULTS: Pathology extraction tasks achieved near-perfect accuracy. For prostate biopsy reports, exact extraction of total cores and involved cores was highly accurate (total cores: accuracy 98.0% [95% CI, 93.0 to 99.5]; involved cores: accuracy 95.0% [95% CI, 88.8 to 98.5]). Performance was similarly strong for RP and TURP reports. On MRI pelvis, extraction of PIRADS scores (accuracy 98.0% [95% CI, 93.0 to 99.5]), lesion locations (accuracy 100% [95% CI, 96.3 to 100]), and lesion dimensions (accuracy 100% [95% CI, 96.3 to 100]) was excellent. For PSMA PET/CT, PPVs were 100% (95% CI, 93.5 to 100) for nodal metastases and 97.9% (95% CI, 89.9 to 99.6) for bone metastases; Tc-99m bone scan performance was comparable. Lower PPVs were observed for nodal and bone metastases on pelvic MRI (84.2%-86.4%) and CT (88.0%-90.3%), largely due to ambiguous language in radiology report texts. CONCLUSION: LLMs can reliably extract key prostate cancer phenotypes from unstructured text across multiple pathology and radiology report types. Ambiguous or indeterminate language remains the principal challenge for optimal performance.

Background: Medications such as proton pump inhibitors (PPIs), antihistamines, and nonsteroidal anti‐inflammatory drugs have been linked to immune checkpoint inhibitor (ICI) efficacy in patients with non–small cell lung cancer (NSCLC), but these associations may reflect unmeasured confounding rather than true pharmacologic effects. This study evaluated whether commonly prescribed medications influence ICI outcomes, using a national patient sample and a negative control cohort. Methods: The authors identified Veterans Health Administration (VHA) patients with stage IV NSCLC treated with first‐ or second‐line ICI therapy ( n = 3739) or chemotherapy ( n = 6585) from 2005 to 2023. Baseline use of 20 common medication classes and an immunomodulatory drug score were assessed. Propensity‐weighted Cox regression evaluated associations between each medication class and overall survival (OS) or time‐to‐next treatment (TTNT) in the ICI group. For any medication with a nominally significant association ( p < .05), the same analysis was repeated in the chemotherapy group to test for nonspecific effects. Results: After propensity weighting, 14 of 20 medication classes showed no association with OS or TTNT in the ICI cohort. Loop diuretics, anticoagulants, opioids, penicillin antibiotics, and fluoroquinolone antibiotics were associated with worse outcomes, but similar effects were seen in the chemotherapy group. A higher immunomodulatory drug score was also associated with inferior outcomes among ICI patients, but this association was likewise present in the chemotherapy cohort. Conclusion: In this study, commonly prescribed medications did not appear to alter ICI efficacy in stage IV NSCLC. Prior associations reported in the literature may be attributable to unmeasured confounding rather than true drug–immunotherapy interactions.

Background: The 2024 American Urological Association, American Society for Radiation Oncology, and Society of Urologic Oncology practice guidelines recommend early salvage radiation therapy (RT) for biochemical recurrence after radical prostatectomy and androgen deprivation therapy for high-risk features. Increasingly, men with high-risk disease are undergoing radical prostatectomy. We therefore characterized contemporary RT and androgen deprivation therapy practices within the Michigan Radiation Oncology Quality Consortium and Michigan Urological Surgery Improvement Collaborative. Methods: Patients receiving postprostatectomy RT from June 9, 2020, to June 9, 2024, were eligible. Prospectively collected data included surgical pathology, RT, and androgen deprivation therapy details. RT was adjuvant (pre-RT prostate-specific antigen [PSA] <0.1 ng/mL), consolidative (persistent PSA ≥0.1), or salvage (all others). Multivariable analyses evaluated associations between clinicopathologic features and androgen deprivation therapy use. Results: Among 345 patients across 26 centers, 56% had at least 1 high-risk feature: pT3b/T4 (24%), pN1 (6%), grade group 4/5 (30%), pre-RT PSA greater than 0.5 ng/mL (27%). RT was adjuvant (10%), consolidative (28%), or salvage (62%), initiated at median PSA of 0.07 ng/mL (interquartile range [IQR] = 0.03-0.09 ng/mL), 0.5 ng/mL (IQR = 0.3-1.5 ng/mL), and 0.3 ng/mL (IQR = 0.2-0.5 ng/mL), respectively. Median time to RT was 8, 6, and 29 months. A minority were recommended 24 months of androgen deprivation therapy (17%), and very few were recommended intensification with AR-pathway inhibitors (5%). On multivariate analysis, androgen deprivation therapy was associated with pT3b/T4 (odds ratio [OR] = 2.77, 95% confidence interval [CI] = 1.34 to 5.93), pN1 (OR = 6.22, 95% CI = 1.35 to 47.57), grade group 4/5 (OR = 2.87, 95% CI = 1.51 to 5.56), and pre-RT PSA more than 0.5 (OR = 2.11, 95% CI = 1.17 to 3.91). Conclusions: Within the Michigan Radiation Oncology Quality Consortium, more than half who received postprostatectomy RT had high-risk features; nearly 30% required consolidation for persistently positive PSA. Androgen deprivation therapy was associated with high-risk features, but few received androgen deprivation therapy prolongation or intensification. Studies are needed to personalize androgen deprivation therapy, especially for those with persistent PSA, who are frequently treated yet underrepresented in trials.

PURPOSE: The 2022 AUA/ASTRO guidelines recommend 18-36 months of androgen deprivation therapy (ADT) with definitive radiotherapy for localized, high-risk prostate cancer. The STAMPEDE M0 trial supports intensification with androgen receptor pathway inhibitors (ARPIs) for patients with ≥2 cT3/T4, Grade Group [GG] 4-5, prostate-specific antigen (PSA) ≥40 ng/mL, or cN1. Given advances in imaging, risk stratification, and treatment delivery, we characterized contemporary practice patterns using prospective data from the Michigan Radiation Oncology Quality Consortium (MROQC). METHODS: Patients enrolled in MROQC with intact, high-risk M0/N0-1 prostate cancer were included. Clinical information, including intended ADT duration and ARPI use, was prospectively collected. The primary outcome was intended guideline-concordant ADT (GC-ADT, ≥18 months). Multivariable analyses (MVA) assessed associations between clinical factors and GC-ADT recommendations. We compared the adoption of ARPI with standard therapies before and after the publication of STAMPEDE M0. Facility-level variability was evaluated using a mixed-effects model, with the treatment site as a random intercept. RESULTS: Between June 2020 and November 2024, 553 patients across 26 centers were included: cT3/4 (13.3%), cN1 (19.9%), GG 4-5 (75.0%), and PSA ≥20 ng/mL (40.0%). Overall, 91.3% were recommended ADT, with 67.0% being guideline-concordant. On MVA, GC-ADT was significantly associated with cN1 (odds ratio [OR], 2.94 [95% CI, 1.44 to 5.99]), GG (GG4 OR, 6.23 [95% CI, 2.85 to 13.62]; GG5 OR, 9.45 [95% CI, 4.46 to 20.06]), and PSA ≥40 (OR, 3.64 [95% CI, 1.22–10.87]). Facility-level variability persisted in the MVA ( P < .0001). Among the 27.9% who met meeting STAMPEDE criteria, ARPI recommendations increased from 0% prepublication to 23.2% afterward. CONCLUSION: Within a statewide quality consortium, guideline-concordant ADT recommendations occurred in two thirds of patients, with ARPI intensification in under 25% among STAMPEDE-eligible patients. These findings highlight the need for individualized ADT strategies and collaborative efforts to standardize high-quality care.

PURPOSE: Women living with HIV (WLWH) experience decreased breast cancer survival. We sought to determine whether WLWH surviving breast cancer also experienced different quality of life (QOL) gain. METHODS: Women who enrolled in the Thabatse Cancer Cohort across oncology centers in Botswana for the initial treatment of stage I-III breast cancer from October 2010 to February 2022 were included. Exclusion criteria were no documented definitive therapy and incomplete data at treatment end or 24 ± 3 months after treatment. QOL was measured quarterly using the SF-8 questionnaire. G methods using weighted exposure and outcome modules were used to mitigate potential bias from imbalances in demographic and cancer characteristics by HIV status. Primary analysis was change in physical component summary (PCS) and mental component summary (MCS) from treatment end to 24 months after treatment for WLWH compared with women without HIV. RESULTS: Of 603 women enrolled, the final analysis included 298, comprising 85 WLWH and 213 women without HIV. Most common reasons for exclusion were no documented definitive treatment (n = 114) and death before 21 months after treatment (n = 137). WLWH were younger, were less wealthy, and had more estrogen receptor/progesterone receptor positive tumors. Overall, PCS and MCS significantly increased from treatment end to 24 months after treatment, from 50.8 to 52.8 and 51.8 to 53.7, respectively. There was no difference in the change of the PCS or MCS with HIV infection, 2.2 (95% CI, –0.4 to 4.9) and 0.6 (95% CI, –1.7 to 2.9), respectively. CONCLUSION: HIV infection did not impede QOL gain at 24 months after treatment in women surviving breast cancer. Further work is needed to clarify the role of HIV on specific treatment-related morbidities and in other malignancies.

The oligometastatic disease state, defined as a cancer with 5 or fewer sites of metastasis, is a therapeutic opportunity to improve oncologic outcomes. Colorectal cancer (CRC) was among the first for which oligometastatic treatment was used in routine clinical practice, and recent studies have shown potential for improved overall survival with metastasis-directed therapies. As CRC is the third most common cause of cancer death in men and women, improving oncologic outcomes in this population is of paramount importance. The relatively recent identification of this treatment paradigm and paucity of high-quality data have led to heterogeneity in clinical practice. This review will explore perspectives of a panel of surgical and radiation oncologists for complex or controversial cases of metastatic CRC.