An approach to estimating prognosis using fractional polynomials in metastatic renal carcinoma

Abstract

We present a prognostic model for metastatic renal cell carcinoma based on fractional polynomials. We retrospectively analysed 425 metastatic renal cell carcinoma patients treated with subcutaneous recombinant cytokine-based home therapies in consecutive trials. In our approach, we categorised a continuous prognostic index produced by the multivariable fractional polynomial (MFP) algorithm, using a strategy in which continuous predictors are kept continuous. The MFP algorithm selected five prognostic factors as significant at the 5% level in a multivariable model: lymph node metastases, liver metastases, bone metastases, age, C-reactive protein and neutrophils. The MFP model allowed us to divide patients into four risk groups achieving median overall survivals of 38 months (low risk), 23 months (low intermediate risk), 15 months (high intermediate risk) and 5.6 months (high risk). Our approach, based on categorising a continuous prognostic index produced by the MFP algorithm, allowed more flexibility in the determination of risk groups than traditional approaches. British Journal of Cancer (2006) For renal cell carcinoma, certain prognostic staging factors, notably, performance status, disease-free interval, erythrocyte sedimentation rate (ESR), lactate dehydrogenase (LDH), neutrophils, haemoglobin, extrapulmonary and bone metastases, and number of metastatic sites were identified as prognostic factors for survival (Elson et al, 1988;Palmer et al, 1992;Lopez-Hänninen et al, 1996;Gelb, 1997;Culine et al, 1998;Hoffmann et al, 1999;Motzer et al, 1999Motzer et al, , 2002Atzpodien et al, 2003).However, the importance of each predictor varies from study to study and is, therefore, controversial. Besides heterogeneity in patient populations and treatments between different studies, a substantial reason for the observed variation might be attributable to an inadequate use of statistical methods (Simon and Altman, 1994).Most researchers who develop and publish prognostic models in cancer seem to assume that to introduce continuous predictors, such as age and haemoglobin, into a multivariable statistical model, it is necessary first to 'categorise' the predictors into two groups. However, the choice of an appropriate cutpoint is not usually obvious a priori. To avert the worry that an arbitrary choice may be sub-optimal, there have been strategies searching for the 'optimal' cutpoint for each continuous predictor, thus, yielding the smallest P-value when testing the effect of the categorised predictor in a univariate Cox model or log-rank analysis. Once such a set of cutpoints has been found, the final multivariable model is often determined by applying a standard algorithm, such as stepwise selection of variables, to the candidate predictors. Sometimes, only those individually significant at the 5% level are considered as candidates for inclusion in the multivariable model.The disadvantages of such a modelling strategy have been rehearsed quite often in the statistical literature. Here, we will illustrate an alternative...

Flexible modeling improves assessment of prognostic value of C-reactive protein in advanced non-small cell lung cancer

“…Overall, our results illustrate potentially serious clinical and research implications of imposing these assumptions a priori, which may lead to a failure to identify important prognostic factors, such as albumin in our study, inaccurate identification of high-risk groups, or spurious contradictions between the results of short-vs long-term prognostic studies. These results are in line with several other clinical and methodological studies indicating important violations of PH and/or linearity hypotheses (Ramsay, 1988;Sleeper and Harrington, 1990;Hastie and Tibshirani, 1990;Royston and Altman, 1994;Greenland, 1995;Rothman et al, 1995;Abrahamowicz et al, 1997;Benedetti and Abrahamowicz, 2004;Royston et al, 2006;Spix et al, 2008). The PH assumption can be tested with simple parametric or non-parametric tests available in a standard statistical software packages (Wei, 1984).…”

Prognostic factors for overall survival with targeted therapy in Chinese patients with metastatic renal cell carcinoma

“…In our study, lung metastasis was found in 61.3% (73/119) patients and was controlled with targeted therapy. With respect to bone metastasis, previous studies have shown mixed results, with bone metastasis identified as a significant prognostic factor in some but not all analyses; 4,15,[21][22][23] however, our study identified bone metastasis as a significant prognostic factor for OS (p < 0.05). The presence of liver or pancreas metastasis was identified as a significant prognostic predictor of shorter OS (p < 0.05).…”

Modeling the relationship between circulating tumour cells number and prognosis of metastatic breast cancer

“…The disadvantages of this approach, related to multiple testing and generalizability of the results, have been abundantly discussed in the literature [15,16,[21][22][23]. Alternative strategies to dichotomization, in which continuous predictors are kept continuous and trends are investigated, have been proposed: among these, the most commonly used are fractional polynomials and spline regression models [23][24][25][26][27][28]. When using the restricted cubic spline regression method [24,25], we observed a significant nonlinear increase in risk of both progression and death with increasing number of CTCs; this increase rate lessened after approximately 5 CTCs.…”