The Molecular Tumor Board Portal supports clinical decisions and automated reporting for precision oncology

Abstract

There is a growing need for systems that efficiently support the work of medical teams at the precision-oncology point of care. Here, we present the implementation of the Molecular Tumor Board Portal (MTBP), an academic clinical decision support system developed under the umbrella of Cancer Core Europe that creates a unified legal, scientific and technological platform to share and harness next-generation sequencing data. Automating the interpretation and reporting of sequencing results decrease the need for time-consuming manual procedures that are prone to errors. The adoption of an expert-agreed process to systematically link tumor molecular profiles with clinical actions promotes consistent decision-making and structured data capture across the connected centers. The use of information-rich patient reports with interactive content facilitates collaborative discussion of complex cases during virtual molecular tumor board meetings. Overall, streamlined digital systems like the MTBP are crucial to better address the challenges brought by precision oncology and accelerate the use of emerging biomarkers.

Fig. 1. Clinical relevance of cancer gene variants.

a, Overview of gene mutations (single-nucleotide variants and small indels) reported as biomarkers of cancer diagnosis, prognosis and/or drug response by three publicly available knowledge bases (CIViC (Clinical Interpretation of Variants in Cancer), OncoKB (Oncology Knowledge Base) and CGI (Cancer Genome Interpreter)^–) at the moment of writing. An assertion corresponds to a reported biomarker effect for a given gene variant, in a given cancer type and with a given level of supporting evidence. Assertions supported by weaker or inconclusive evidence (as provided by the knowledge base metadata when appropriate; Methods) are excluded from these results. 1,000g, 1000 Genomes Project; AF, allele frequency. b, Representation of distinct levels of interpretation for cancer gene variants. The functional relevance evaluates the allele-centric effect of the observed variant, whereas context-dependent interpretation factors in additional considerations (such as whether the variant is germline or somatic, co-occurring alterations in the same or other genes and/or the cancer type of the patient’s tumor). Questions that are addressed in each step are exemplified here for a given BRCA1 mutation (HRR, homologous recombination repair; PARPi, poly-ADP ribose polymerase inhibitors). Source data

Fig. 2. Interpretation of cancer gene variants.

a, The MTBP classifies a given cancer gene variant as (putative) functionally relevant or neutral according to three distinct sources of evidence (named A, B and C here) or of unknown relevance if none of these criteria are fulfilled. Note that the knowledge bases listed here are those integrated at the moment of writing^–, but their usage may be subject to changes depending on evolving needs and preferences. FI, functional impact; OncoKB-mut and OncoKB-biom refer to the biological and predictive relevance annotation of variants in OncoKB, respectively. b, Criteria supporting the variant functional classification are considered to provide strong (>0.9 certainty) or very strong (>0.99 certainty) evidence as extrapolated from the work in variant pathogenicity classification, following the rationale described in the table. c, Aggregated knowledge base assertions (excluding those from genetics population data) at the moment of writing. As expected by the different scopes of each knowledge base and the long tail of lowly recurrent mutations, only a minority of the variants appear curated in more than one knowledge base, which stresses the importance of their aggregation to provide a comprehensive annotation. d, Graphical summary of some of the criteria used for assuming that a variant with null consequence type is disrupting the function of a given tumor suppressor (part of the evidence of type B; a). These are largely based on established rules to identify loss-of-function variants in Mendelian genes (Methods). e, The lowest level of evidence to estimate a given variant effect is based on bioinformatics metrics. For variants that are not located in mutation hotspots, we decided to use the combined annotation dependent depletion (CADD) score to estimate the functional relevance of missense mutations in tumor suppressor genes (TSGs), as functional impact predictions perform worse in other scenarios (data not shown). The method and associated thresholds were selected according to our own benchmarking, based on the performance observed for mutations with curated effects (upper violin plot) and in silico simulations (lower violin plot) (Methods). FN, false negative (given these thresholds); FP, false positive; TN, true negative; TP, true positive. Source data

Fig. 3. MTBP functional classification of variants in the Cancer Core Europe cohort.

a, Functional classification of the tumor mutations observed in the CCE cohort (variants assumed to be functionally neutral, such as common polymorphisms, are not included here). Colors represent the different sources of supporting evidence used by the MTBP (Fig. 2). The upper pie chart represents overall counts of mutations, whereas the lower pie chart represents counts of unique mutations. b, Functional classification of the tumor mutations (represented as in a) observed across several of the most recurrently mutated cancer genes in the CCE cohort. Tumor suppressors tend to accumulate a variety of null events (e.g., frameshift or nonsense variants) whose effect is mainly estimated by bona fide biological assumptions, except for genes that have been more exhaustively characterized by ongoing efforts (e.g., BRCA1 and BRCA2) and/or are enriched by specific dominant-negative mutations (e.g., TP53 and, to a lesser extent, PTEN). In oncogenes, variants are concentrated in few hotspots with a well-known gain-of-function consequence in some genes (e.g., KRAS and BRAF), whereas others show a diversity of mutations whose effect remains unclassified in a considerable number of cases (e.g., ERBB2 and FGFR3). MoA, mechanism of action. Source data

Fig. 4. MTBP cancer biomarkers matching in the Cancer Core Europe cohort.

a, The MTBP matches the observed tumor variants with cancer biomarkers reported as a specific nucleotide or protein change (see examples next to the red vertical lines), a categorical genomic definition (red horizontal line examples) or a functional entity (matched according to the MTBP functional interpretation; red circle examples). Variant actionability must also factor in (among others) the coincidence between the biomarker and patient’s cancer type; the MTBP takes into account the disease hierarchy so the biomarker is matched when reported for the patient’s tumor type or a subtype thereof (red arrow). b, The MTBP ranks the cancer biomarkers (diagnosis, prognosis and drug response) found in the tumor following the ESMO/ESCAT scale with two minor modifications (Methods). An approximate equivalence with the actionability proposed by the Association for Molecular Pathology, American Society of Clinical Oncology and College of American Pathologists (AMP/ASCO/CAP) is also shown (left). The table summarizes how the MTBP classifies the biomarker actionability according to the coincidence of the (i) variant and (ii) cancer type reported for the patient’s tumor versus biomarker and (iii) the clinical evidence supporting the biomarker effect (as curated by each of the biomarker knowledge bases used by the MTBP at the moment of writing). c, Tumor mutations observed in the CCE cohort matching with drug sensitivity biomarkers (biomarkers of resistance or toxicity are not included here). Colors represent the highest level of actionability of all the biomarkers found in each tumor. Numbers above the bars represent the median (percentile 25–75) number of unique mutations per tumor reported as drug sensitivity biomarkers, regardless of their level of actionability. Cancer types are grouped by their tissue of origin. Source data

Fig. 5. Use of the MTBP in Cancer Core Europe.

a, Patients of the CCE cohort recommended for one of the Basket of Baskets trial arms available at the moment of the molecular tumor board discussion. The pie chart shows the proportion of these patients that were not enrolled due to clinical deterioration and/or screening failures. For the remaining cases, other investigational treatments were evaluated in the respective patient’s medical institution without requiring CCE consensus (data not shown). b, Patients in which the MTBP issued an alert due to germline variant(s) associated with inherited cancer risk (restricted to the 375 patients with sequencing of both tumor and normal-paired samples). The pie chart shows the proportion of these cases that would have been overlooked by current germline testing guidelines, as variant carriers did not meet personal criteria. Bar plots detail the genes in which these pathogenic variants were identified and whether they are currently described as increasing the risk for the respective patient’s index cancer type. c, The MTBP reports used in CCE contain an integrated dashboard with clinical and pathology information, details of the sequencing assay(s) and a version control of the resources used to annotate the data. Clinically relevant molecular signatures are listed together with individual gene alterations, the latter organized in three different tables according to their functional classification (functionally relevant, unclassified and functionally neutral). These tables summarize the evidence supporting the variants’ functional classification and their associated actionability, including in-house clinical initiatives (prioritizing eligibility for clinical trials available across the CCE network) and matching with cancer biomarkers reported to date (diagnosis, prognosis and drug response, tiered according to the ESMO-ESCAT scale; see Fig. 4). Access to detailed information is provided by the use of interactive elements in the HTML report, as exemplified here by the pop-up windows opened in 1–4. d, The lines represent the time (median and IQR) devoted to discussing each patient’s case during the CCE virtual molecular tumor board meetings. The circles indicate the accumulated number of reviewed cases (overall 500 patients). BoB, Basket of Baskets; VAF, variant allele frequency. Source data

Extended Data Fig. 1. Public version of the Molecular Tumor Board Portal.

a, The public MTBP is an open resource for research purposes that provides a lightweight version of the MTBP analytical pipeline. This public resource is accessed via a website (https://www.mtbp.org/) that offers an interface to upload the gene variants to analyze. At the moment of writing this manuscript, the public MTBP only supports the analysis of single-nucleotide variants and small indels. b, The public version of the MTBP does not issue actionability flags that require information that cannot be inferred from the generic input employed here and/or require interpretation nuances that may differ across investigators/institutions (such as matching with a given portfolio of clinical trials or the identification of germline incidental findings). Instead, the public MTBP provides a general interpretation of the functional and predictive relevance of the uploaded variants, with the aim of supporting a detailed review of the user according to his/her specific needs. c, Public MTBP website activity from the date of its release until the moment of writing this manuscript.