Reanalysis of Clinical Exome Sequencing Data

The rapid accrual of knowledge in genomic medicine has prompted the reanalysis of pre-existing data.^1,2 We clinically reanalyzed data from two patient series that had undergone diagnostic proband-only exome sequencing.

The exome sequences of the first series of 250 patients were obtained between 2011 and 2012.³ Cumulative reanalyses of extant data after the release of the initial clinical report increased the molecular diagnostic yield of cases from 24.8% to 46.8% (comprehensive manual reanalysis performed based on knowledge source from 12/2017) (Figure 1). Although clinical laboratories often use a “manual” approach (see Supplementary Appendix) for routine exome-sequence analysis, this approach may not be sustainable at scale, particularly for iterative re-analyses accruing historical cases. We designed a semi-automated reanalysis process, which prioritizes variants based on genetic and/or functional evidence and genes based on semantic phenotypic similarity scores (Supplemental Appendix). This approach, when used to analyze the first series, achieved a diagnostic sensitivity of 92.9% (with the manual reanalysis as the reference standard), while sustaining a manageable workload under a model of annual reanalysis over a five-year period (Supplemental Appendix). This semi-automated reanalysis approach was then applied on a second cohort of 2000 consecutive cases originally analyzed in 2012–2013,⁴ augmenting the molecular diagnostic rate from 25.2% to 36.7% (Figure 1). We did not carry out manual reanalysis of the exome sequences of the second, much larger cohort, given the burden of the task. Potential factors contributing to the difference in the overall diagnostic rates between the two series (46.8% vs. 36.7%) include different percentages of patients with Mendelian disorders and a difference in sensitivity between comprehensive manual and semi-automated analysis.

Figure 1. Systematic reanalysis increases diagnostic rates in two patient series referred to clinical exome sequencing.

The left (Cohort #1, N=250) and the right (Cohort #2, N=2000) panels depict the molecular diagnostic rate changes from the original rates, evidenced by respective publications^3,4, and the reanalysis rates. The yield alteration originates from patients whose diagnostic molecular findings changed from zero to one or more (new diagnosis, highlighted in red in the reanalysis bar), from one to two or more (multilocus pathogenic variation; the white segment below the red segment in the reanalysis bar), and from one to zero (overturned molecular diagnosis, see Supplemental Appendix). The white segment above the red segment represents the proportion of patients who received a partial diagnosis from reanalysis (n=2 for Cohort #1, n=5 for Cohort #2), which is not counted towards the diagnostic rate. The Pareto charts to the right of the bar charts illustrate the breakdown of the interpretive reasons contributing to each new molecular diagnosis, with cumulative counts from each category shown in the bottom axis and the cumulative percentages shown in the top axis.

In line with the rapid pace of improvement in knowledge of genetic causes of disease, the vast majority of new molecular diagnoses resulted from newly discovered disease genes (75% and 64% from Cohorts #1 and #2, respectively; Supplemental Appendix, Figure S1). Additionally, several newly implicated genes initially associated with childhood-onset neurologic disorders, including PURA, DDX3X, and TANGO2, rank highly amongst the most frequently mutated in our clinical cohorts (Fig. S2 in the Supplementary Appendix). New molecular diagnoses also resulted from upgraded variant classifications in known disease genes (5.9% and 14% from Cohorts #1 and #2, respectively; Supplemental Appendix, Table S2). Other contributors include the emergence of additional clinically observed phenotypes, the identification of pathogenic copy-number variants, and the positive identification of a genetic cause that had been previously missed (Supplemental Appendix). The number of patients with multiple molecular diagnoses increased from 25 to 48 (in the combined series) after reanalysis, illustrating the contribution of continued genomic analyses to the deconvolution of clinically blended phenotypes (Supplemental Appendix, Table S5).⁵

Education is necessary for physicians and patients to understand the concept of an ‘evolving’ molecular interpretation. The communication and counseling processes are complicated pragmatically by the temporal separation between the initial and the updated reports and the dilemma of whom to send the report to. We sent a survey to 55 healthcare providers of the 64 patients in the first series for whom the reanalysis had yielded new molecular diagnoses. We received a response from 23 (42%) providers regarding follow-up data for 42 (66%) patients. Of these 23 providers, 10 reported a preference that all physicians involved in the patient’s care be informed of the results of the reanalysis, and 13 reported a preference that only the physician who ordered the reanalysis or who had ordered the original analysis be informed. The respondents reported that about three quarters of their patients (30) had received genetic counseling for the updated findings, and of these 30, about half (17) had their clinical management changed as a consequence of the new results (Supplemental Appendix, Table S7). The 12 patients who did not receive genetic counseling about their new diagnoses were reported to have relocated, died, or received their new diagnosis but did not keep their follow-up appointment. These findings suggest that periodic reanalysis may benefit patients and their families and physicians, and that they should be counselled about this at the time that clinical exome sequencing is ordered. However, an important caveat to our conclusions is that our study was not designed to measure clinical benefit of the intervention.