After the establishment of the initial models, the same team worked on performing temporal validation of the models by testing them on patients attending the same clinic in a different year (Chow et al. 2009). This work utilized a third form of test called the D index to assess the performance of the models. The D index quantified the discrimination of the models based on their abilities to separate the risk of death among groups of patients; for this statistical test, the larger the D statistic, the greater the degree of separation. The models had C index values ranging from 0.65 to 0.68, which represented reasonable concordance, and D index values of 0.99 to 1.41, representing good separation of prognostic groups. However, generalized R² values remained poor (range: 0.27–0.31), indicating that the models accounted for less than one-third of the variability in survival.

Chow and colleagues aimed to simplify their original prognostication models (Chow et al. 2008). They cited difficulty in extracting ESAS scores as the reason why they believed a simpler model made up of only three readily available factors—primary cancer site (breast, prostate, lung and others), site of metastases (bone vs. others) and KPS—deserved consideration. As with their previous models, they utilized two scoring approaches: the partial score method and the number of risk factors method. In order to validate these new models, they conducted both temporal and external validation by comparing cohorts from different years and at different institutions. In evaluation of the performance of these predictive models, C index values ranged from 0.63 to 0.66 and D index values ranged from 0.81 to 1.09, indicating reasonable separation of the three prognostic groups formed by the models. In addition, it was noted that the models were not hindered by over-optimism. In contrast to the previous R² values for the initial (R² = 0.31) and temporal validation (R² = 0.27) cohorts of the six-factor models, the R² values for the three-variable models were lower across the initial, temporal and external validation sets (R² = 0.23, 0.24, 0.15, respectively). The reduction in the generalized R² values indicated that the three ESAS variables that were omitted did contribute statistical significance to the six-factor model. However, these statistics should be considered with reservation, as R² values are generally not reported in predictive models, and when they are reported, values are generally low despite successful validation (Toscani and Brunelli 2005).

Although some factors were not included in the models by Chow et al., in general, survival of metastatic cancer patients is dependent on performance status (Chow et al. 2002), symptoms comprising part of the “terminal cancer syndrome” (Reuben et al. 1988), and other patient symptoms which include asthenia, anorexia, cachexia, weight loss, dysphagia, xerostomia, delirium and cognitive impairment (Llobera et al. 2000; Maltoni et al. 2005). Additionally, some studies state that nonclinical factors, such as social support (Schoenbach et al. 1986) and religion (Phillips and Smith 1990; Jarvis and Northcott 1987) also influence survival in this patient population, although others have concluded that psychosocial factors are not associated with survival in palliative advanced cancer patients (Chow et al. 2008; Chow et al. 2002).