TABLE 1.
Technical aspects of the CDSS projects.
| CDSS name | Context of usage | Input data | Algorithm technology | Reference data | Individual assessment: prediction outcome | Predictive performance metric | Year of first report |
|---|---|---|---|---|---|---|---|
| AEDSS | Assessment during monitoring | Functional system scores |
Expert system: production rules, extended Backus–Naur form syntax More recently: ontology (Operational Conceptual Modeling Language) |
Expert knowledge (Kurtze rules) | EDSS | Kappa coefficient | 2002 |
| EBDiMS |
Prognostic counseling Initial prognosis |
Short‐term version: MS course, age at onset, EDSS, disease duration, number of relapses in the past 12 months Long‐term version: MS course, gender, age at onset, number of attacks in the first 2 years |
Data visualization: weighted distance‐based patient matching; statistical analysis in R | SLCMSR database reusing RCT and OS data; 45 sources | Individual risk profile: EDSS course, time to sustained progression, time to EDSS = 6 | Brier Score | 2007 |
| MS Prediction Score | Transition from RR‐MS to SP‐MS | Age, time since the last relapse, type of the last relapse, remission from the latest relapse | Model‐based: continuous hazard functions estimated by Poisson regression models |
Gothenberg incidental cohort Uppsala cohort from the SMSreg |
Yearly probability of transition to SP‐MS | Inferential model | 2014 |
| MS BioScreen | Personalized monitoring | Any biomarker collected in the MS EPIC cohort | Data visualization: patients matching algorithm with customizable variables | MS EPIC cohort at the University of California, San Francisco | Overlay of the patient's trajectory with the distribution of similar patients for any biomarker in the reference database | None | 2014 |
| MS Prognosis Simulation | First relapse, first year of RR‐MS, monitoring the consecutive years | Age at onset of disease, gender, sphincter onset, pure motor onset, motor–sensory onset, sequel after onset, number of involved functional systems at onset, number of sphincter plus motor relapses, EDSS ≥ 4 | Expert system: agent‐based modeling, NetLogo 5.0.4 | Aggregated results in literature and 50 patients from the Hospital Egas Moniz, Lisboa, Portugal | Conversion to CDMS at 10 and 20 years, conversion to SP‐MS at 10 years, risk of reaching EDSS = 6 at 10 and 20 years | Pearson correlation coefficient | 2014 |
| Function Watch (SMSreg) | Assessment during monitoring | Gender, age, disease duration, treatment; maximum 2‐year‐old reference data | Data visualization: patients matching algorithm with customizable variables | Sweden MS Registry | Function Watch diagram: overlay of the patient's metrics with the distribution of similar patients for EDSS, MSSS, MSFC, MSIS‐29‐Physical, MSIS‐29‐Psychological, SDMT, FSMC, FSS, EQ5D, activity/work capacity, MS‐checklist, SF36‐1 | None | 2015 |
| Bloodwatch, RiskMx | Alemtuzumab treatment monitoring | Laboratory results (in HL7 format) | Expert system: RiskMxTM system, matching against monitoring schedule and reference ranges | Laboratory reference ranges | Alert the patient and neurologist in case of abnormal value or missed blood draw | None | 2019 |
| Clinical Decision Support System for Multiple Sclerosis Diagnosis | Diagnosis of RR‐MS | 45 demographic, clinical, and paraclinical items | Expert system: production rules |
Decision tables, decision trees, and semantic networks according to the 2004 MS Diagnosis Guideline and McDonald's 2017 diagnostic criteria 130 medical records from the Shahid Beheshti Hospital of Kashan, Iran |
Diagnosis of RR‐MS | AUC and kappa coefficient | 2020 |
| MSProDiscuss | Transition from RR‐MS to SP‐MS | Multiple demographic, clinical, and paraclinical items | Model‐based: scoring algorithm, multiple logistic regression | Observational study of 3294 MS patients in the USA | Likelihood of progression | Inferential model with subsequent ranking and weighting of the predictors by physicians | 2020 |
| MS TreatSim, UISS‐MS | Initial prognosis | Presence of oligoclonal bands, age, lesion load, treatment | Model‐based: agent‐based modeling, Protégé OWL, UML modeling | Aggregated data from MS literature, AFFIRM trial dataset | Relapses (as oligodendrocyte loss), cytokines, and immune cell population dynamics | Statistical comparison of the in silico results versus the real results | 2020 |
| PHREND (DESTINY) | Treatment selection | Age, gender, EDSS, index treatment, past treatment, disease duration, time since last relapse, relapse count, DMT count, efficacy class of the past treatment, duration of the past treatment, duration of the index treatment, and clinical site | Model‐based: hierarchical Bayesian generalized linear model |
NeuroTransData MS registry, CONFIRM, DEFINE, REGARD, TRANSFORMS, AFFIRM, CLARITY, OPERA I/II, and TEMSO trials. |
Number of relapses, progression‐free MRI, and confirmed disability progression up to 4.5 years | Mean squared error, negative log‐likelihood, and Harrell's concordance statistic (C‐index) | 2020 |
| Prognosis for patients with RR‐MS a | Prognosis counseling | Age, gender, disease duration, EDSS, number of GdE lesions, number of previous relapses during the previous 2 years, months since the last relapse, whether it is on treatment | Hierarchical Bayesian generalized linear model | Swiss Multiple Sclerosis Cohort | Relapses at 2 years | C‐statistic | 2021 |
| MS Vista (PRIMUS) | Treatment selection | Age, sex, age at onset, disease duration, MS type, EDSS at the last visit, number of relapses within the past 12 months, number of T2 lesions on current MRI, GdE lesions on current MRI, number of new T2 lesions within the past 12 months | Data visualization: filter‐based patient matching algorithm | ADVANCE trial dataset | Relapses, new MRI lesions, confirmed disability progression at 1 year | None | 2022 |
| sNfL reference app a |
Assessment during monitoring Assessment of treatment response |
Age, body mass index, sNfL level | Model‐based: GAMLSS | Swiss Multiple Sclerosis Cohort, SMSreg, normative dataset of 4532 persons in the USA | Age‐ and body mass index‐adjusted sNfL percentile values and z‐scores | Odds ratio | 2022 |
Note: Platform names are mentioned in parentheses.
Abbreviations: AEDSS, Automatic Expanded Disability Status Scale; AUC, area under the curve; CDMS, clinically definite MS; CDSS, clinical decision support system; DMT, disease‐modifying treatment; EBDiMS, Evidence‐Based Decision Support Tool in Multiple Sclerosis; EDSS, Expanded Disability Status Scale; EPIC, Epigenetics, Proteomics, Imaging, Clinical; EQ5D, EuroQoL group health questionnaire; FSMC, Fatigue Scale for Motor and Cognitive Functions; FSS, Fatigue Severity Scale; GAMLSS, generalized additive model for location, scale, and shape; GdE, gadolinium‐enhancing; MRI, magnetic resonance imaging; MS, multiple sclerosis; MSFC, Multiple Sclerosis Functional Composite; MSIS‐29, Multiple Sclerosis Impact Scale–29; MSSS, Multiple Sclerosis Severity Scale; OS, observational study; OWL, Web Ontology Language; PHREND, Predictive Healthcare with Real‐World Evidence for Neurological Disorders; PRIMUS, Projections in Multiple Sclerosis; RCT, randomized clinical trial; RR‐MS, relapsing–remitting MS; SDMT, Symbol Digit Modalities Test; SF36‐1, the first question from the SF‐36 Health Survey; SLCMSR, Sylvia Lawry Centre for MS Research; SMSreg, Swedish MS Registry; sNfL, serum neurofilament light chain; SP‐MS, secondary progressive MS; UISS‐MS, Universal Immune System Simulator (MS extension); UML, unified modeling language.
The “Prognosis for patients with RR‐MS” and “sNfL reference app” names were given by the authors of this review. Inferential models are modeling approaches that do not assess predictive performance metrics, like for a classification or regression task.