Development of Malaysia breast cancer survival prognostic tool (myBeST) for prediction of survival probability among women with breast cancer in Malaysia
Background: Breast cancer accounts for a sizeable portion of newly diagnosed cancer. Prognostic tools were developed to inform patients regarding their outcomes. Performance of Western-centric tools found to be less accurate when applied in our setting with PREDICT breast cancer (PREDICT) had an...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2023
|
| Subjects: | |
| Online Access: | http://eprints.usm.my/59062/1/Mohd%20Nasrullah-24%20pages.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Background: Breast cancer accounts for a sizeable portion of newly diagnosed cancer.
Prognostic tools were developed to inform patients regarding their outcomes.
Performance of Western-centric tools found to be less accurate when applied in our
setting with PREDICT breast cancer (PREDICT) had an acceptable accuracy.
Objective: The study aimed to develop predictive models for survival among women
with breast cancer in Malaysia, to compare its performance with PREDICT and the
model’s algorithm was incorporated to develop a web-based Malaysian Breast Cancer
Survival Prognostic Tool (myBeST).
Methodology: This study consists of two phases. Phase 1 is a retrospective cohort
study using data abstracted from seven regional breast cancer referral centres in
Malaysia. We collected 13 predictors and survival outcomes. Time-to-event Cox
proportional hazard (PH) analysis and two supervised machine learning classifiers
(decision tree (DT) and artificial neural networks (ANN)) were employed to model
and predict five-year survival probability. The model with the best performance
indices was compared with the PREDICT tool. Subsequently, in Phase 2, the model
was deployed in a web-based format with accompanying content to describe the tool.
The website underwent several user-centred iterative development stages, including content (n = 8) and face validity (n = 20) assessments by medical specialists and
medical officers.
Results: There were 1,006 patients included for model derivation and validation. They
were mostly Malay, with ductal carcinoma, hormone-sensitive, HER2-negative, at T2,
N1-stage, without metastasis, received surgery and chemotherapy. The five-year
survival was 60.5% (95% CI: 57.6, 63.6). By the Cox PH model, Indians had a higher
hazard of death compared to Malay (Adjusted HR (Adj. HR): 1.77, 95% CI: 1.19,
2.63). Histological type, cancer grade, tumour, node, and metastasis stage at diagnosis
significantly associated with death. Those who received surgery (Adj. HR: 0.49, 95%
CI: 0.28, 0.87), chemotherapy (Adj. HR: 0.59, 95% CI: 0.44, 0.79), and radiotherapy
(Adj. HR: 0.70, 95% CI: 0.51, 0.96) had a lower risk of death. Cox PH model
outperformed the DT and ANN model in terms of accuracy (Cox PH: 0.841, DT:
0.811, ANN: 0.821), F1-score (Cox PH: 0.879, DT: 0.859, ANN: 0.870) and the area
under the receiver operating characteristic curve (AUC; Cox PH: 0.891, DT: 0.39,
ANN: 0.877). The Cox PH was more accurate in predicting five-year survival
probability with a higher AUC (0.78, 95% CI: 0.73, 0.82) than PREDICT (AUC: 0.75,
95% CI 0.70, 0.80). Thus, the model was deployed as the main feature of our webbased
prognostic tool. The website was developed and improved at every iterative
stage. The content validity indices were ≥0.88 and face validity indices were >0.90,
resulting in a functioning and user-centred prognostic tool.
Conclusion: The web-based tool derived from robust Cox PH model showed
promising results. Further validation, usability, and feasibility studies are necessary as
the tool could potentially be used by care providers to convey individualised survival
prediction for newly diagnosed breast cancer patients. |
|---|