Solid organ transplantation can enable long-term survival for patients with end-stage organ disease. To prevent rejection, recipients must take immunosuppressive medications that increase the risk of cancer (1, 2) and may also promote the recurrence of cancer among individuals with a prior cancer diagnosis (3, 4).

The decision regarding whether a patient can be placed on the transplant waitlist takes account of comorbid medical conditions that would make transplantation at high risk for failure or of limited benefit (5, 6). Most organ transplants are not performed as treatment for cancer, but a prior cancer diagnosis is a common comorbidity among transplant candidates and recipients. Moreover, a prior cancer diagnosis is associated with increased posttransplant mortality related, in part, to cancer recurrence (7, 8). Because many cancers recur in the first few years after diagnosis, guidelines for evaluating transplant candidates with a history of cancer typically recommend a waiting period that depends on the type of cancer and stage at diagnosis (9–13). A “5-year cancer survival rate of near 80%” is used as a benchmark for proceeding to transplantation in recent guidelines (13).

For patients with a history of cancer, the selection process is intended to identify those individuals for transplantation who have an acceptably low risk of cancer recurrence and death from their cancer (i.e., cancer-specific mortality). This approach corresponds to identifying patients who have a high likelihood of being cured of their cancer, which can be modeled statistically as a cure probability (14–16). For a population of cancer patients, the cure probability is defined as the proportion of people who will not die from their cancer no matter how long they are followed.

Using data from the Transplant Cancer Match (TCM) Study, we recently demonstrated that estimated cure probabilities provide a useful metric to assess prognosis in cancer patients undergoing organ transplantation (17). Our study evaluated 5425 cancer patients with 17 different types of malignancy who underwent solid organ transplantation in the United States (US). These cancer patients’ cure probabilities were typically very high at the time of transplantation, reflecting a good cancer prognosis due to careful selection. Importantly, however, we found that those patients whose cure probability was in the lowest third (i.e., tertile) had increased cancer-specific mortality following transplantation (adjusted hazard ratio [HR] 2.08, 95% confidence interval [95%CI] 1.48–2.93, vs. the highest tertile).

Our prior study is an important proof-of-principle of the potential utility of cure models for candidate selection for transplantation. However, it had several limitations. First, because we included 17 different cancer types, the statistical models were not tailored to any single cancer, complicating their application to individual patients. Second, we did not attempt to validate the models or recommend a clinically useful threshold for decision-making. Third, we did not provide a tool for clinicians to estimate an individual patient’s prognosis.

Colorectal cancer is a common malignancy among people considered for organ transplantation. In the present study, we developed and validated statistical cure models tailored to colorectal cancer and applied them to US patients undergoing solid organ transplantation during 1991–2019. Based on these models, we provide estimates that a colorectal cancer patient will not die from their cancer in the next five years, calculated both at the time of their cancer diagnosis and updated to accommodate the passage of time since their cancer diagnosis. We illustrate how these probabilities can inform guidelines for transplant candidate selection. In addition, we developed a web-based tool based on these models that clinicians can use to calculate probabilities for individual patients at any time at or after their cancer diagnosis to assist in transplant evaluations.

In the present study, we developed and validated statistical cure models for patients with colorectal cancer. The distribution of cure probabilities for colorectal cancer patients in the US general population was distinctly bimodal, corresponding to their differing prognoses based on demographic and clinical factors. The cancer-specific survival probabilities generated from these models were predictive of subsequent mortality from cancer, both among cancer patients in the general population and among patients transplanted during 1991–2019.

Cure models are an attractive framework for assessing prognosis of cancer patients (14–16) and have intuitive appeal in the setting of evaluating of organ transplantation candidates (17). If a patient can be considered likely to be cured of their cancer (i.e., that the cancer will not lead to death), then it seems reasonable to discount the prior cancer history when evaluating the candidate for transplantation. In this study, we used the cure models to derive 5yCSS probabilities, with the idea that these would be readily interpretable by clinicians. Offering transplantation to a patient with a high risk of cancer recurrence (i.e., low 5yCSS) might be considered inappropriate, both from the perspective of the patient and with respect to responsible stewardship of the scarce supply of donor organs.

Several characteristics of the colorectal cancer patients who received an organ transplant after their cancer diagnosis reflect careful selection by transplant providers (Table 1). They were younger than patients in the general population and, notably, almost all had been diagnosed with localized or regional stage cancer. Over a median period of 6.3 years between cancer diagnosis and transplantation, the 5yCSS probabilities increased modestly for most patients, but some patients had large increases during this interval (Figure 1D). As a result, at the time of transplantation, the median estimated 5yCSS was 0.96, and 87.7% of transplanted patients had 5yCSS probabilities of at least 0.90. The long average time interval between cancer diagnosis and transplantation partly reflects caution among providers, as the interval was longer for patients with regionally advanced or metastatic cancer (Supplemental Table 3). Another reason for the long time intervals is that some patients were diagnosed with cancer before developing end-stage organ disease.

Due to careful selection by clinicians, very few recipients in our study died from their cancer after transplantation. Nonetheless, cancer-specific mortality was more than three-fold higher among recipients whose 5yCSS was less than 0.90 at the time of transplant (HR 3.31, 95%CI 1.52–7.21). If our results based on transplants during 1991–2019 can be extrapolated to the present and future, they suggest that a threshold 5yCSS of 0.90 could be appropriate to consider in offering a transplant to an individual with a history of colorectal cancer. The characteristics of the minority of transplanted patients with an estimated 5yCSS below 0.90 reflect a combination of adverse clinical characteristics (e.g., advanced cancer stage, short time from diagnosis to transplantation) as well as countervailing influences on clinicians’ decisions to proceed with transplantation even in the presence of a relatively poor cancer prognosis (e.g., young patient age, transplantation of an organ other than kidney) (Supplemental Table 2A).

Because of their utility in predicting cancer-specific mortality, these models can inform interpretation and possibly lead to revision of guidelines regarding which patients should be considered eligible for transplantation (Table 3). For some situations, our results suggest that wait times might be shortened. For example, Al-Adra et al. recommend a waiting time of 2 years after completion of cancer treatment for patients with AJCC6 stage IIA colon cancer (13). In contrast, our results indicate that transplantation may be safe much earlier, because such patients (like those with stage I colon cancer) already have a median 5yCSS above 0.90 at the time of cancer diagnosis.

Nonetheless, while our results suggest that decreases in wait times for some patients may be justified, it is important to remain cautious. Mortality is high among cancer patients (including those with colorectal cancer) who are already immunosuppressed at the time of their cancer diagnosis (18–20), suggesting that intact immunity is especially important in controlling cancer at the time of initial cancer diagnosis and treatment. Thus, if transplantation were offered too soon after a cancer diagnosis, immunosuppression could potentially promote cancer progression or cause relapse even among patients with very good prognoses.

In other scenarios, the wait times may need to be longer than currently recommended. In particular, Al-Adra et al. recommend that patients with stage IIB, IIIB, and IIIC colorectal cancer should be considered eligible for transplantation at 3–5 years after completion of cancer treatment (13). As shown in Table 3, this range of wait times may be reasonable if 0.80 is the threshold 5yCSS for offering a transplant but may be insufficient if the threshold is 0.90. Finally, the median 5yCSS for patients with stage IV colorectal cancer remains quite low 5 years after diagnosis (Table 3), emphasizing that offering transplantation to such individuals (with the possible exception of liver transplantation for patients with isolated liver metastases) (13, 21, 22) should be undertaken with the understanding that there is a high risk of relapse.

Although the median 5yCSS estimated for patients at specified times since cancer diagnosis (Table 2) is informative, there is substantial variability among patients based on their demographic and treatment characteristics (Figure 3, Supplemental Figure 3). We therefore developed a web-based tool (description with link: https://dceg.cancer.gov/tools/risk-assessment/calculator-of-colorectal-cancer-survival-probability; direct link: https://appshare.cancer.gov/crcprobability) that allows providers to calculate 5yCSS individualized for a specific patient at the time of evaluation for listing or transplantation. This individualized estimate could be especially useful in the transplant setting when the range of 5yCSS probabilities in the general population data crosses 0.80 or 0.90 (Figure 3). The tool also provides the estimated 5yCSS at different timepoints after cancer diagnosis, allowing clinicians to determine the potential benefit of waiting longer before offering a transplant.

Importantly, we did not have data to fit cure models specifically for cancer patients with end-stage organ disease, although their 5yCSS probabilities might differ from those predicted by our models. Our models also do not include information on molecular tumor characteristics, details of cancer treatment (which might differ between transplanted patients and others in the general population), or updated information available to clinicians about the status of the patient’s cancer at the time of evaluation. Thus, the individual patient probabilities provided by our web-based tool should be viewed as a rough guide for the provider’s own estimate of the patient’s 5yCSS probability. The individualized 5yCSS estimate would have to be lowered if the clinician needed to account for additional adverse patient characteristics or, obviously, clinical or laboratory evidence for cancer recurrence. Correspondingly, the provider could increase the estimated 5yCSS if there was no evidence for disease after a thorough assessment.

We note several study limitations. First, the cure models were developed using US data, so they may not generalize to other countries. Second, many transplanted patients developed cancer before 2004, so AJCC6 stage was frequently missing in this group. Nonetheless, SEER historic stage was available for many more patients, and we showed that probabilities using SEER historic and AJCC6 stages were highly correlated. Furthermore, SEER historic stage is a simple classification scheme that could still be utilized for calculating 5yCSS probabilities even after AJCC staging is updated. Third, imaging and cancer staging procedures are continually improving. This effect, over time, generally “upstages” cancers, when, for example, an increased sensitivity of imaging reclassifies patients from non-metastatic to metastatic disease. Finally, cancer care is continually improving, and a survival probability estimated for a patient in the past is likely lower than for a similar patient treated now and in the future; thus, our 5yCSS estimates likely represent conservative (i.e., worst-case) scenarios for current and future patients. Future iterations of our models will be needed to reflect changes in diagnosis and treatment.

This study has multiple strengths, including that the cancer patients were systematically identified from large population-based databases and were representative of US cancer patients and transplant recipients. We used cross-validation to internally validate our cure models. Also, our estimated 5yCSS probabilities at diagnosis are similar to 5-year relative survival estimates provided by the American Cancer Society (see Table 2 vs. relative survival of 91%, 72%, and 13%, respectively, for localized, regional, and distant stage colon cancer) (https://www.cancer.org/cancer/types/colon-rectal-cancer/detection-diagnosis-staging/survival-rates.html). Finally, in Table 3 we provide suggested wait times based on 5yCSS thresholds of 0.80 (as proposed by al-Adra et al. (13)) or 0.90. Nonetheless, our webtool allows for more flexible transplant provider decision-making for an individual patient and varying risk threshold.

We focused on colorectal cancer because it is a common malignancy among both transplant candidates and the general population. This or similar approaches could be used in the future for other cancers provided that sufficient data from cancer registries or other large representative sources are available. A challenge for some cancers is that important prognostic information (e.g., Breslow depth for melanoma) may be difficult to obtain. Another difficulty for some cancers is the need to incorporate updated clinical information after cancer diagnosis on routinely obtained measures of residual disease (i.e., the prognosis of transplanted patients will be much better than for patients in the general population if screening during transplant evaluation has ruled out residual cancer).

In conclusion, we believe that cure models are a useful framework for developing wait time guidelines and evaluating individual cancer patients for organ transplantation. The complex decision to offer transplantation may be improved by incorporating cancer prognostic information throughout the evaluation process.