1430P - Optimal cut-points for QLQ-C30 scales associated with overall survival in patients with advanced hepatocellular carcinoma (AHCC): a comparison of tw...

Date 01 October 2012
Event ESMO Congress 2012
Session Poster presentation III
Topics Supportive Measures
Hepatobiliary Cancers
Presenter Momar Diouf
Authors M. Diouf1, F. Bonnetain2, J. Barbare1, O. Bouché3, J. Meynier1, L. Dahan4, X. Paoletti5, T. Filleron6
  • 1Clinical Research, Amiens University Hospital, 80000 - Amiens/FR
  • 2Biostatistic And Epidemiological Unit(ea 4184), Centre Georges François Leclerc, 21000 - Dijon/FR
  • 3Hopital Robert Debré, Reims/FR
  • 45service D’hépato-gastroentérologie Et Oncologie Digestive, AP-HM, 13000 - Marseille/FR
  • 5Service De Biostatistique, Institut Curie, 75005 - Paris/FR
  • 6Institut Claudius Regaud, 31052 - Toulouse/FR



Health-related quality of life (QoL) has been validated as prognostic factor for patients with aHCC. However, to be used in routine practice, QoL should be dichotomized. Usually, cut-points were based on arbitrary percentile value. The main objective of this study was to identify optimal cut-offs for 5 scales: global health (GH), physical functioning (PF), role functioning (RF), fatigue (FA) and diarrhea (DIA). Two published methods were used to compare distributions of cut-offs and their risk-ratio. We finally evaluated the improvement of existing prognostic classifications (PC) by dichotomized QoL.

Patients and methods

271 patients with aHCC were included in CHOC trial. QoL was assessed in the 2 weeks prior to randomization with the EORTC QLQ-C30. Identification of optimal cut-points was based on two univariate methods proposed by Mazumdar and Farragi respectively. Mazumdar method was based on the « minimum p-value » approach. Adjustment of type I error were based on Altman formula. For Farragi method, the total sample was divided into two sub-samples: learning and validation. A cut-point was derived for each sub-sample using « minimal p-value » and each patient was classified according to the cut-point for the sub-sample to which it does not belong. The final cut-point was the one that minimize the p-value in the total sample. Stability of the results was evaluated using boostrap procedure (n = 500). Improvement of PC was studied with multivariate Cox model, QoL being dichotomized at its optimal cut-point.


QoL was available in 234 patients (86%). For RF, the most frequent cut-point was 30 with the two methods (Mazumdar: 496/500 – Farragi: 500/500). Univariate HR (95%CI) were 2.99 [1.62 – 5.52] and 2.80 [2.58 – 3.04] respectively for Mazumdar and Farragi. In Farragi method, the 2 sub-samples found the same cut-offs in 486/500B. The recommended cut-points were respectively 45, 50, 30, 30 and 5 for GH, PF, FA, RF and DIA. Compared to CLIP + WHO PS, CLIP + QoL + clinical factors rose the C-index from 0.65 [0.62 – 0.69] to 0.70 [0.66 – 0.74].


The stability of the cut-points was good and precision of CI acceptable for both methods, but better for Farragi. Interestingly, this categorization of QoL increased the performance of all PC and should be considered as competing factor to WHO performance status.


All authors have declared no conflicts of interest.