Development of a Score Predicting Survival after Palliative Reirradiation

Purpose. To develop a prognostic model for predicting survival after palliative reirradiation (PR). Methods and Materials. We analyzed all 87 PR courses administered at a dedicated palliative radiotherapy facility between 20.06.2007 (opening) and 31.12.2009. Uni- and multivariate survival analyses were performed, the previously published survival prediction score (SPS) was evaluated, and a PR-specific prognostic score was calculated. Results. In multivariate analysis, four parameters significantly influenced survival: performance status, use of steroids, presence of liver metastases, and pleural effusion. Based on these parameters, a 4-tiered score was developed. Median survival was 24.5 months for the favorable group, 9.7 and 2.8 months for the two intermediate groups, and 1.1 months for the unfavorable group (P = 0.019 for comparison between the two favorable groups and P ≤ 0.002 for all other pair-wise comparisons). All patients in the unfavorable group died within 2 months. Conclusion. The performance of PR-specific score was promising and might facilitate identification of patients who survive long enough to benefit from PR. It should be validated in independent patient groups, ideally from several institutions and countries.


Introduction
Palliative reirradiation is currently used in a variety of clinical settings, including but not limited to bone and brain metastases or lung and gynecological cancers [1][2][3][4]. The number of scientific publications on this topic has increased in recent years [5]. In a well-defined geographical part of Norway, palliative reirradiation contributed 10% to all palliative radiotherapy series administered during a 12-month period [6]. Randomized trials comparing single versus multiple fractions for painful bone metastases reported retreatment rates of 11-42% after a single fraction and 0-24% after multiple fractions, as summarized by Chow et al. [1]. Comparable to palliative radiotherapy in general, clinicians attempt to tailor treatment regimens to patients' prognosis, thereby minimizing undesirable over-and undertreatment. Decision aids such as prognostic scores and nomograms might facilitate rapid and reproducible assessment of patients' survival expectation by transforming the complex set of patient-and disease-related prognostic factors into a standardized tool. Ideally, prognostic scores are easy to administer and valid across different institutions and countries [7]. The Survival Prediction Score (SPS), developed and validated by Chow et al. in patient cohorts treated with palliative radiotherapy, is among the tools that might be widely applicable, because it is based on three readily available parameters: primary cancer type, site of metastases, and performance status [8]. Its performance has never been tested specifically in patients undergoing palliative reirradiation. Together with a large number of other baseline factors potentially impacting survival, we analyzed SPS in a single-institution cohort study.

Methods
We retrospectively reviewed the records of all consecutive patients who received palliative reirradiation at a single 2 Journal of Oncology hospital with dedicated palliative radiotherapy unit. The patients started their treatment in the time period from June 20, 2007 (date of opening of the dedicated palliative radiotherapy unit), to December 31, 2009. Reirradiation was defined as partial or complete field overlap (examples of partial overlap: initial course included thoracic vertebrae Th4-6 and reirradiation Th6-8; initial course of radical prostate radiotherapy followed by pelvic bone metastasis irradiation). A total of 87 reirradiation courses were studied. Stereotactic radiotherapy was unavailable and thus not included in the present series. All medical records, treatment details, and information on date of death were available in the hospital's electronic patient record (EPR) system. The survival status and date of death or last follow-up of the patients were obtained from the EPR. Patients who were lost to followup were censored on the date of last documented contact (personal appointment, telephone conversation, and blood test). Median follow-up for all surviving/censored patients was 5.4 months. Survival time was measured from the start of reirradiation. Actuarial survival curves were generated by Kaplan-Meier method and compared by log-rank test (analyses performed with IBM SPSS Statistics 20). Multivariate analyses were performed by Cox regression (backward conditional method). We assigned SPS as described by Chow et al. [8], that is, based on three variables (nonbreast cancer, metastases other than bone, and Karnofsky performance status (KPS) ≤ 60): poor prognosis group when all three were present, intermediate prognosis group when two were present, and good prognosis group when 0-1 were present. Our own prognostic scores were developed as previously described by Rades et al. [9,10]. In brief, the score for each predictive factor was determined by dividing the actuarial death rate at prespecified time points (given as the percentage) by 10. For example, patients with good KPS were assigned 0 points and those with poor KPS 1.5 points (rate of death at 1 month (15%) divided by 10). The total score represented the sum of the scores for each predictive factor. Two time points reflecting poor prognosis or short survival were chosen, 1 month and 2 months, because there is no generally agreed definition of sufficient survival expectation, justifying initiation of palliative radiotherapy. Given that recent research and discussions focused on overtreatment, for example, use of radiation therapy in the last 30 days of life, we felt that predicting short survival might be more important [11][12][13][14].

Results
Median age at the time of reirradiation was 67 years (range 38-90 years). Prostate (29%) and non-small cell lung cancer (NSCLC, 11%) were the most common primary tumors. Additional baseline information is shown in Table 1. Bone metastases were the prevailing target for reirradiation. The most common regime consisted of 10 fractions of 3 Gy (43%). Other common regimes included 8 Gy single fraction (uncomplicated bone metastases) and 5 fractions of 4 Gy (various sites and indications). Five courses (6%) remained incomplete, typically because of earlier than expected clinical We analyzed the potential prognostic impact of all baseline parameters shown in Table 1 and assigned SPS score. However, the performance of this score was unsatisfactory because two of the three patient groups had similar survival ( Figure 2). As shown in Table 2, two components of the SPS score (metastases location and performance status) significantly influenced survival, while primary tumor type  Journal of Oncology did not. In multivariate analysis, a total of four parameters significantly influenced survival: KPS, use of steroids, presence of liver metastases, and pleural effusion. Based on these parameters, a new 4-tiered prognostic score was developed. As described in Section 2, we compared two different variants, which are shown in Table 3. When applying a short-survival-definition of 1 month (variant 1), the resulting survival curves separated clearly ( Figure 3). Median survival was 24.5 months for the favorable group, 9.7 and 2.8 months for the intermediate groups, and 1.1 months for the unfavorable group ( = 0.024 for comparison between the two favorable groups and ≤ 0.003 for all other pair-wise comparisons). Thirty-three percent of patients in the unfavorable group died within 1 month and all within 2 months. When applying a short-survival-definition of 2 months (variant 2), the resulting survival curves separated equally clear (Figure 4). Median survival was exactly the same as in variant 1 ( = 0.019 for comparison between the two favorable groups and ≤ 0.002 for all other pair-wise comparisons). Since the unfavorable group included exactly the same patients, 33% died within 1 month and all within 2 months. Because of its superior significance level, variant 2 might be the preferred assignment method.

Discussion
Palliative reirradiation is an important treatment option, providing symptom improvement in many patients with bone metastases [1] and other conditions [15]. While most previous studies were small and often retrospective, the randomized bone metastases study by Chow et al. comparing different fractionation regimens included 850 patients [1]. Median survival in the two arms was 9.3 and 9.7 months, respectively. This result is comparable to the 8.6 months reported in our own, bone metastases-dominated study. However, survival of individual patients might be as short as few days or as long as several years (Figure 1). Therefore, prognostic scores might be valuable decision aids when prescribing palliative reirradiation. Chow       395 patients referred to their palliative radiotherapy program [16]. Later, they refined their original six-parameter-model by reducing the number of variables to three (primary cancer type, site of metastases, and performance status), arriving at the SPS [8,17]. We hypothesized that this score might also predict survival of patients receiving reirradiation but discovered that further studies, which also include other models, are needed. The performance of the SPS score ( Figure 2) can be explained by the fact that not all adverse SPS features (nonbreast cancer, metastases other than bone, and poor performance status) influenced prognosis of reirradiated patients. In the present study, metastases location and performance status significantly influenced survival, while primary tumor type did not. Disadvantages of our study include its retrospective design and limited number of patients, especially regarding subgroups. Not all patients had complete information on all baseline parameters recorded in the EPR system. The majority of reirradiation courses consisted of hypofractionated regimens, mostly 1-15 fractions, with dose/fractionation parameters reflecting a patient's expected prognosis (clinical estimate). We did not use any particular prognostic models or scores when assigning treatment regime during the time period covered in our study. Nevertheless, more than 90% of patients who were offered reirradiation also completed their treatment. Only 6% were treated during the final month of life, suggesting that our clinical decision making was largely successful, even if further improvement should be attempted.
Our score based on KPS, use of steroids, presence of liver metastases, and pleural effusion performed promisingly. To the best of our knowledge, no other scores related specifically to palliative reirradiation exist. One of the clinical aims of applying prognostic scores might be avoidance of overtreatment in patients with very short survival [18]. Recently, Tanvetyanon et al. have reported on use of radiotherapy in the last 30 days of life in the United States [19]. They used a SEER-Medicare linked database to obtain a large study cohort of 202,299 patients who died as a result of lung, breast, prostate, colorectal, and pancreas cancers (top five cancer causes of death) between January 1, 2000, and December 31, 2007. The rate of radiotherapy in the last 30 days of life, by many regarded as inappropriate overtreatment, though this point of view is controversial, was 7.6%. No data on reirradiation were reported in this study, and no attempt was made to develop predictive models. Before our new score can be widely implemented, external validation is necessary. In the future, it might become possible to study narrowly defined patient groups, if sufficiently large databases can be created. For example, Tanvetyanon et al. have published prognostic factors for survival after salvage reirradiation in patients with head and neck cancer [19]. Rades et al. have developed scores specific to metastatic spinal cord compression [20,21], and Sperduto et al. to brain metastases [22], both related to first line treatment rather than reirradiation.

Conclusions
Prognostic factors for survival might change during the course of disease, for example, from first line to subsequent treatments. The performance of the newly developed score was promising and might facilitate identification of patients who survive long enough to benefit from palliative reirradiation. It should be validated in independent patient groups, ideally from several institutions and countries.