Our model predicted that for men with locally advanced PC who undergo radical prostatectomy, initial observation with early SRT (PSA ≤ 0.5 ng/mL) is slightly less effective than ART for 10-year PSA recurrence-free survival, metastasis-free survival, and overall survival. However, observation with early SRT confers similar or slightly better health-related quality of life to ART; 28% of men similar to those in the SWOG study avoid RT through 10 years and many more delay RT, thus avoiding or delaying the adverse effects of RT. Our model underestimates the real disutility of RT; we assumed RT toxicity to be grade 1 or 2, whereas in the randomized trials grade 3 toxicity occurred in 4%.3,6
Furthermore, assessment of the late effects of RT was only available through 5 years.3,6
Including delayed complications would make ART less attractive.
Our model successfully projected the outcomes of PSA recurrence, metastasis, and death as reported for the SWOG randomized trial of ART versus observation. An obstacle to modeling the SWOG study results, particularly for overall survival, was that the background death rate in SWOG's observation group appeared to be higher than in the ART group. Specifically, the number needed to treat to prevent one metastasis was 12.2 men, but the number needed to treat to prevent one death was only 9.1 men.9
If the survival benefit of ART was achieved by reducing the number of metastases and thus cancer-specific deaths, then one would expect the number needed to treat to prevent one death to be more than
the number needed to treat to prevent one metastasis. This raises the question of whether the observed survival benefit in the ART group was at least partially due to lower non-cancer-specific (i.e., background) death, despite randomization. We were able to account for this difference in survival in our model by either assigning a 67% annual probability of progression from metastasis to cancer death or a 1.25-fold higher background death rate in observation patients. Because a 67% annual death probability from metastatic PC is extremely high, we opted to use the higher background death rate for model calibration, then used equal background death rates for the cohort simulation. For all analyses, we applied a 24% probability of cancer death from metastases in both groups.
Decision models are limited by the assumptions used to build them. In general, ours is a conservative model, with assumptions constructed to bias against observation and SRT. Still, our model has limitations. We assumed that men on observation received SRT at earliest evidence of PSA recurrence (≤ 0.5 ng/mL). In reality, there will be variation in follow-up, and some men may progress beyond the point at which they would maximally benefit from SRT.4
Decisions about a patient's commitment to intense surveillance may need to be made on an individual basis. While our results were robust when assuming that 75% of the men needing SRT actually received it, when that number dropped to 50%, ART was preferred. If >75% of men with a PSA recurrence on observation can be expected to complete SRT then observation becomes an even more attractive option. Such a high SRT capture rate might be achievable in a trial setting; however, in community-based registries the frequency of SRT in men with a PSA recurrence has been as low as 11% in the early 1990s and as high as 32% in 2000–0428,29
. Thus, the ability to follow an observation protocol and willingness to proceed with SRT should be integral to selecting observation vs. ART.
There are limits to the external validity of the results. First, roughly a third of the men in the SWOG and EORTC trials had a persistently elevated PSA post-prostatectomy, which could represent residual cancer. Second, contemporary men have on average lower grade, stage and PSA than men in the era of the SWOG and EORTC trials.2
Although, the more contemporary GCS trial in which all men had a negative post-operative PSA still had a probability of PSA recurrence similar to that in the SWOG and EORTC trials, the GCS men had high risk cancer (8% had T4 disease). A modern patient with only one of the features of locally advanced PC (ECE, PSA or SVI) will typically have a 10-year PSA recurrence-free survival of 70–80%, compared to the 28% used in this model. Third, men in the SWOG trial received SRT at higher PSA thresholds than those used currently. In summary, all three of these features that contrast with current practice (lower risk patients, a negative post-operative PSA and earlier SRT delivery) would only lend stronger support for initial observation in the modern era.
Other limitations bear mention. The negative impact of RT on QALYs significantly impacts the interpretation of our model. As utilities are subjective, future investigations should incorporate sensitivity analyses around the utility values in the model. Our findings should only be applied to men with disease similar to the average participant in the reference cohorts, and not men with metastatic disease, T4 disease, or PSA ≥0.4 post-prostatectomy. In subset analyses, the benefits of adjuvant and SRT were experienced by select high-risk patients;9,17
however, the reference studies lacked sufficient detail to design a model that accommodates factors predictive of individual response to therapy. We model results for a cohort of 65 year old men. When varying the background death rate down to that of a 55 year old man or up to that of a 75 year old man, QALYs remained higher with observation than ART. In sensitivity analysis we demonstrate that ART is preferred over observation for the outcome of 10-year QALYs only when the utility of RT is nearly equal to the utility of observation (0.85 vs. 0.86, respectively). In practical terms this means that ART would be preferred over observation only if radiation was essentially side effect free or if men placed little value on the additional risk of those radiation side effects. Indeed, modern radiation techniques may yield different results from those seen in this study; specifically, rectal complications are lower but urinary complications may be higher with intensity modulated radiation therapy than 3-D conformal radiotherapy.30
While our sensitivity analyses show our findings to be robust over a wide range of ART and SRT effectiveness, the optimum model would allow the clinician to input an individual's disease characteristics and arrive at a preferred treatment strategy. A combination of our model with input probabilities from nomograms such as the Kattan nomogram is intriguing area for future research.2
In our model, ART showed slightly better clinical outcomes compared with observation plus SRT in terms of post-RT PSA recurrence-free survival, metastasis-free survival, and overall survival. When the adverse effects of RT were included in the decision model, observation plus SRT became the preferred treatment. These results were highly dependent on delivering SRT early (at PSA ≤0.5ng/mL) and in ≥75% of observed patients with a PSA recurrence. This model can inform physician-patient treatment discussions as we await the results of randomized clinical studies.