Page 93 - Predicting survival in patients with spinal bone metastasesL

Table 5 shows the c-statistic for each model, and the percentage of patients that were stratified correctly at four, six, twelve and twenty-four months. At four and six months, survival data from patients classified in the worst prognostic group of each model were used (e.g. Tomita category ‘Terminal’ or Bauer category ‘Supportive care’) and at twelve and twenty-four months survival data from patients classified in the best prognostic group of each model were used (e.g. Tomita category ‘Long- term’ or Bauer category ‘Local control’).
Table 5. Model performance. Four and six months based on worst prognostic group, twelve and twenty-four months based on best prognostic group.
SCORING SYSTEMS
Model
% dead within
% alive after
c-statistic
4 months
6 months
12 months
24 months
Tokuhashi Bauer
Tomita
Van der Linden Rades
Bollen (all data) Bollen (external data)
0.64 62% 74% 0.64 64% 76% 0.64 65% 78% 0.66 64% 77% 0.44 69% 81% 0.70 75% 86% 0.69 67% 83%
61% 39% 58% 35% 68% 47% 80% 60% 37% 22% 80% 60% 82% 65%
The Tokuhashi, Bauer and Tomita models have an identical c-statistic and a relatively low predictive ability, both for their worst and best prognostic groups. The Van der Linden model has a similar c-statistic, but is much better at identifying patients with a longer survival. The Rades model has the lowest c-statistic, but shows relatively good predictive power for patients with short survival. However, patients with a long survival are poorly identified. The Bollen model has the highest c-statistic and is good at identifying patients with short survival and patients with long survival. When the results from the external data analysis are observed, the c-statistic remains almost identical. However, the number of patients stratified correctly at four months drops considerably.
DISCUSSION
In this multicenter retrospective study of 1379 patients treated for symptomatic SBM it is demonstrated how predictive models perform on a large independent dataset.
VI
91