Recommendations for future prognostic studies
| Approaches . | Topic . | Area . | Type . | Specifics . |
|---|---|---|---|---|
| To do | Statistical approaches | Model development and validation | Methods used | C-statistic/AUC |
| Positive and negative predictive values | ||||
| Brier scores | ||||
| Net Reclassification Index | ||||
| New investigation areas | Novel prognostic factors | Patient factors | Micro-RNAs | |
| SNPs | ||||
| Organ-specific biomarkers | ||||
| Sarcopenia/muscle mass | ||||
| Validated predictive and brief battery of geriatric assessment tools | ||||
| Newly designed models | Patient factors | CHARM | ||
| Disease factors | DRI updated with molecular and MRD data | |||
| Transplant factors | Appraisal of complexity of different donor types and HLA-matching degrees | |||
| Novel statistical methods | Cubic splines | |||
| Machine learning and artificial intelligence | The least absolute shrinkage and selection operator and object-oriented regression FIS GBM Bayesian belief networks Markov models | |||
| Principal component analysis and joint decomposition regression | ||||
| Novel methodological approaches | Reversibility of prognostic factors | Designing dynamic models that provide different prognostic estimates depending on timing in patient’s treatment journey | ||
| Decision curve analysis | Net benefit evaluations for prediction models | |||
| New specialty | Prognostication | Oncology—palliative | Trained investigators/MDs | |
| Not to do | Validation | Studies | Model performances | Comparing models from different prognostic areas |
| Small sample studies | ||||
| Different diagnoses | ||||
| Different transplant settings | ||||
| Practice | Clinical | Counseling | Ignoring prognostic data | |
| Using physician perception alone | ||||
| Unilateral decision in complicated situations |
| Approaches . | Topic . | Area . | Type . | Specifics . |
|---|---|---|---|---|
| To do | Statistical approaches | Model development and validation | Methods used | C-statistic/AUC |
| Positive and negative predictive values | ||||
| Brier scores | ||||
| Net Reclassification Index | ||||
| New investigation areas | Novel prognostic factors | Patient factors | Micro-RNAs | |
| SNPs | ||||
| Organ-specific biomarkers | ||||
| Sarcopenia/muscle mass | ||||
| Validated predictive and brief battery of geriatric assessment tools | ||||
| Newly designed models | Patient factors | CHARM | ||
| Disease factors | DRI updated with molecular and MRD data | |||
| Transplant factors | Appraisal of complexity of different donor types and HLA-matching degrees | |||
| Novel statistical methods | Cubic splines | |||
| Machine learning and artificial intelligence | The least absolute shrinkage and selection operator and object-oriented regression FIS GBM Bayesian belief networks Markov models | |||
| Principal component analysis and joint decomposition regression | ||||
| Novel methodological approaches | Reversibility of prognostic factors | Designing dynamic models that provide different prognostic estimates depending on timing in patient’s treatment journey | ||
| Decision curve analysis | Net benefit evaluations for prediction models | |||
| New specialty | Prognostication | Oncology—palliative | Trained investigators/MDs | |
| Not to do | Validation | Studies | Model performances | Comparing models from different prognostic areas |
| Small sample studies | ||||
| Different diagnoses | ||||
| Different transplant settings | ||||
| Practice | Clinical | Counseling | Ignoring prognostic data | |
| Using physician perception alone | ||||
| Unilateral decision in complicated situations |
CHARM, comprehensive health assessment risk model; FIS, fuzzy inference system; GBM; gradient boosting machine; SNPs, single nucleotide polymorphisms.