Statistical aspects of prognostic factor studies in oncology

Simon, Richard; Altman, Douglas G.

doi:10.1038/bjc.1994.192

Cited by 533 publications

(349 citation statements)

References 36 publications

Supporting

Mentioning

335

Contrasting

Unclassified

Order By: Relevance

“…The multiple comparisons-corrected P values accompanying their addition were only 0.073 (9 ϫ 0.0081) and 0.387 (9 ϫ 0.043). 33 In view of the disadvantages of frequent changes in prognostic indices, the evidence for doing so should be really convincing. Therefore, we choose not to do so at this moment.…”

Section: Statistical Validationmentioning

confidence: 99%

Prognostic index for patients with parotid carcinoma

Poorten

Hart

Laan

et al. 2003

Cancer

View full text Add to dashboard Cite

BACKGROUNDValidation of the prognostic indices for the recurrence‐free interval of patients with parotid carcinoma, the development of which was described in a previous report, is needed to be confident of their generalizability and justified prospective use.METHODSThe Dutch Cooperative Group on Head and Neck Cancer database contains 231 patients with parotid carcinoma from six tertiary referral centers who were seen between 1985–1994. This database was used to validate the predictive value of the prognostic indices, PS1 (pretreatment index predicting recurrence) and PS2 (posttreatment index predicting recurrence), in patients with parotid carcinoma. Validation methods included calculation of both indices for each patient, comparison of coefficients, construction of survival curves using the published cutoff points, and calculation of concordance measure C. Wald tests for optimization of scale and weights of the contributing variables, and for possible score improvement by including other variables, were also performed.RESULTSThe 5‐year disease‐free rate was 62% (standard error = 4%). PS1 was calculated for 183 patients. The previously set cutoff points resulted in 5‐year disease‐free rates that ranged from 92% (PS1 = 1) to 42% for the least favorable group (PS1 = 4). Concordance measure C was 0.74. The postoperative score, PS2, was calculated for 171 patients. Previous cutoff points resulted in 5‐year disease‐free rates that ranged from 90% (PS2 = 1) to 40% (PS2 = 4). Concordance measure C was 0.71. Both PS1 and PS2 did not improve using the findings in this independent material.CONCLUSIONSThe prognostic indices performed adequately in this validation sample. This finding demonstrates that the prognosis of a parotid carcinoma patient can be quantified by using a weighted combination of the parameters of age, pain, clinical T (cT) classification, clinical N (cN) classification, skin invasion, facial nerve dysfunction, perineural growth, and involved surgical margins. A quantified prognosis, in terms of the recurrence‐free interval, can be used to provide information to the patient, to perform clinical trials, and eventually to make clinical decisions. A user‐friendly translation and computerized calculation of the indices should be the next step toward generalized prospective use and repetitive evaluation of the indices. Cancer 2003;97:1453–63. © 2003 American Cancer Society.DOI 10.1002/cncr.11254

show abstract

Section: Statistical Validationmentioning

confidence: 99%

Prognostic index for patients with parotid carcinoma

Poorten

Hart

Laan

et al. 2003

Cancer

View full text Add to dashboard Cite

show abstract

“…Conflicting results, however, were reported in the literature and this might be due to small sample sizes, nonhomogeneous populations (Simon and Altman, 1994) and the use of optimal cut-off values . Also, various techniques were employed without proper methodological evaluation (Press et al, 1994).…”

mentioning

confidence: 92%

Multiple molecular marker testing (p53, C-Ki-ras, c-erbB-2) improves estimation of prognosis in potentially curative resected non-small cell lung cancer

et al. 2000

View full text Add to dashboard Cite

Summary A prospective study was performed in patients with non-small cell lung cancer (NSCLC) to evaluate the prognostic importance of multiple molecular marker (p53, c-Ki-ras, c-erbB-2) testing. 103 patients with potentially curative resections (RO resection) for NSCLC in histopathological stages I-IIIA were included. SSCP analysis and DNA sequencing for p53 and c-Ki-ras genes were performed on paired tumour and normal lung tissue samples and immunohistochemistry (c-erbB-2) was done on frozen tissue sections with a specific anti-c-erbB-2 monoclonal antibody. 46/103 (44.6%) NSCLC showed p53 mutations and 17/103 (16.5%) c-Ki-ras mutations including 12/37 (32.4%) adenocarcinomas. Overexpression of c-erbB-2 (p185) was detected in 56/103 (54.4%) tumours. 24/103 (23.3%) NSCLC were negative for alterations in all 3 parameters (c-Ki-ras, p53 and p185) whereas 79/103 (76.7%) were positive for at least one of the 3 parameters. In a regression model including a multiple molecular marker parameter (negative for all 3 markers versus positive for at least one marker), histopathological stage (P < 0.00001), respectively the pT (P < 0.01) and pN (P < 0.00001) categories and the multiple molecular marker parameter (P < 0.01) were of significant prognostic importance. This study demonstrates that testing 3 molecular markers (c-Ki-ras, p53 and c-erbB-2) improves estimation of prognosis compared to single marker testing and appears to define low (82.6% ± 7.9% 5-year survival) and high risk (40.2% ± 5.5% 5-year survival) groups for treatment failure in potentially curative (RO) resected NSCLC.

show abstract

“…Regulatory agencies now require that biomarker cut-off points splitting patients into high and low risk groups be defined and validated for use in patient populations (51). Splitting a continuous measure into a dichotomous group reduces the power to detect a real association with outcome, but if optimized cut-off values are used, then they should be determined using a training data set with an independent testing data set to validate the cut-off point (51).…”

Section: Discussionmentioning

confidence: 99%

“…Using a cut-off point as reported from another study or defined based on the distribution of marker level among patients without use of clinical outcome data is also unbiased (51).…”

Section: Discussionmentioning

confidence: 99%

“…Additionally, it needs to have therapeutic implications readily interpretable by a clinician and have been validated in independently confirmed phase III studies (51). Biologically, this requires that the marker can be assessed reproducibly in numerous clinical samples by a well-characterized, standardised, accurate and quality controlled assay system (50).…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation