الفهرس 
Review of LiteratureOnly 14 pages are availabe for public view
 |  | from | 115 |  |  |
| | | from | 115 | | |
Abstract
B
one metastasis is a complex condition with uncertain survival outcomes. To select patients who may benefit from surgery, accurate estimation of preoperative survival is crucial.
Various prognostic models, including scoring models and machine learning algorithms, have been developed and externally validated.
This study aimed to provide a comprehensive review of survival prediction scores for patients with spinal and extremity metastases. The meta-analysis included 35 articles and revealed diagnostic odds ratios and area under the curve values for different scoring systems.
The primary cancer site plays a significant role in survival prediction. The study identified the SORG Nomogram and SORG SGB machine learning algorithm as reliable options for estimating survival in patients with spinal metastases. These models showed good discriminatory ability and calibration. The Nomogram incorporates relevant prognostic factors and has a user-friendly format.
The study also evaluated survival prediction models for extremity metastasis disease and found that the SPRING 2013 nomogram and PATHfx 3.0 model demonstrated superior performance in terms of discrimination and accuracy. However, consensus on the variables to be included in these models is lacking.
Factors such as primary tumor histology, impending and pathologic fractures, and laboratory markers have been identified as important predictors of survival. The study emphasized the need for methodological guidelines, prospective validation, and comparison of scoring systems.
Individual risk prediction models are gaining popularity due to their superior performance. However, even the best-performing models still have room for improvement. Collaboration among healthcare professionals and standardized treatment regimens are essential for making therapeutic decisions.
In our study, the SORG Nomogram and machine learning algorithms showed proficiency in predicting survival outcomes for surgical interventions in patients with spinal metastases. These models can provide valuable information for treatment planning.
For extremity metastasis disease, the PATHFx 3.0, 2013-SPRING, and possibly OptiModel models demonstrated the best performance in terms of both discrimination accuracy and calibration scores.
To ensure the widespread use of a prediction algorithm, it is important to externally validate it in different populations. The PATHfx model stands out as it addresses the issue of missing data and is clinically user-friendly, while also being widely externally validated.
Conclusion
Individualized preoperative survival prediction models for patients with MBD are a valuable decision-making tool for orthopedic surgeons, but there has been limited improvement in their calibration accuracy and discriminatory power.
Currently, it is challenging to recommend a specific model as the standard for predicting survival in MBD patients, but our results demonstrated that the SORG Nomogram and machine learning algorithms exhibit proficiency in predicting survival outcomes for surgical interventions targeting spinal metastases.
Further, our results showed that PATHFx 3.0, 2013-SPRING, and possibly OptiModel appear to be the best models in terms of both discrimination accuracy and calibration scores for extremity metastasis disease.
Widespread use of a prediction algorithm requires external validation in different populations, and only the PATHfx model addresses the issue of missing data while also being clinically user-friendly and widely externally validated.
Limitations
There were some limitations to our study, including inter-study heterogeneity, the lack of consideration of treatment effects on overall survival and cause of death, and the limited availability of individual data from enrolled studies.
The use of different statistical methods for performance evaluation, inclusion of diverse scoring systems, and the use of various time points have led to heterogeneity, which posed some difficulties. However, we addressed these limitations by focusing on comparable metrics and presenting the results in a comprehensive format. Most comparative studies have primarily assessed discrimination, while calibration has not received enough attention.
It is challenging to eliminate the effects of primary tumor types in current studies.
We only included comparative validation studies to capture superior findings, and we did not cover all external validation studies on individual scores. Nonetheless, based on our knowledge, we consider it unlikely that other scoring systems would perform better than those mentioned above. Lastly, our conclusions can only be generalized to surgical cohorts.
The retrospective design of many studies and external validation cohort analyses. This design does not allow for uniformity in diagnostics and operative treatment criteria, leading to possible differences in local treatments between centers and over time.
Additionally, most of the reviewed models had limitations due to missing data or dropped variables. To overcome this, some studies used an interpretation of the clinical description, while others dropped variables with missing data, which could compromise the discriminatory abilities of the models.
The Bayesian network, a machine learning method used by Pathfx 1.0 and Pathfx 3.0, performed well despite differences in patient populations and varying amounts of incomplete or missing data in several external validations, indicating its ability to effectively account for uncertainty within the data.
Recommendations
We recommend the following:
Future studies should explore new prognostic factors and develop or edit current scoring systems using machine learning algorithms.
Consistent external validation and comparison of these systems in large prospective cohorts are essential to understand their capabilities, weaknesses, and net benefits fully.
Interdisciplinary collaboration among oncologists, radiologists, and orthopedic surgeons is needed to make the final therapeutic decision rather than relying solely on risk prediction models.
It also encourages future prospective multicenter studies to investigate and compare the most promising models using standardized treatment regimes.
Scoring systems are only a more objective aid in surgical decision-making and should not be used as an absolute tool with fixed thresholds.
Decision-making regarding surgery in patients with spinal metastasis should be based on a multidisciplinary approach, rather than relying solely on Tokuhashi and Tomita scores. These scores could be incorporated as part of a larger decision-making process.
Decision for or against surgery should never be based alone on a prognostic score but should take symptoms like pain or neurological compromise into account.
We recommend future large, multicenter, prospective studies to compare the PATHfx 3.0, SPRING 2013, and OptiModel using the same external validation dataset