Targeted High-throughput Sequencing for Hematological Malignancies: A GBMHM Survey of Practice and Cost Evaluation in France

Targeted NGS panels

Diagnostic NGS strategies vary significantly between centers. Some laboratories use a number of specific panels whereas others used composite panels, reporting on only a subset of the genes tested, according to the type of hematological malignancy. Details of all panels used during this period (2018–2022) are not provided because their precise content evolved during the impact evaluation period, particularly for capture panels. Some laboratories choose to transmit information to prescribers on all abnormalities considered to be significant in all targeted genes/hot-spots, while others only report information relevant to the disease category, prescription request, and/or protocol agreements regarding actionable targets, while maintaining the possibility of more extensive reporting, when necessary.

Group of molecular biologists for hematological malignancies

The GBMHM was heavily inspired by the European initiatives such as the European Society for Laboratory Hematology-Oncology (ESLHO), https://eslho.org and the European LeukemiaNet, https://www.leukemia-net.org/home/. The GBMHM technological platform is accredited as a provider of external quality assessment (EQA) by the HAS CA. Combined education and bi-annual EQA by the onco-hematology diagnostic community have allowed collective, nationwide, improvement in standardization and harmonized reporting of both CE-marked and, predominantly, in-house IVD testing.^3.

Impact analysis

To describe the expected impact of NGS molecular diagnosis on clinical decisions and care circuits, a prospective, one-arm multicentric observational study was conducted. Participating centers are all GBMHM members and each undertook to report a predefined number of consecutive NGS prescriptions, depending on their anticipated activity during the evaluation period. There was no a priori sample size calculation for this observational study.

Only clinical NGS, defined as prescribed tests that led to an individual result transmitted to the prescriber and recorded in the patient file with an identifiable time delay between prescription and reporting, was to be included. Research activity, leading uniquely to the deposition of results in clinical trial databases but not communicated to prescribers, was not collected.

Only an expected impact on the decision and care circuit could be measured in the absence of record linkage that would allow full individual patient follow-up. It was, therefore, assumed that the NGS result would be impactful, based on the data collected by the platforms if (1) an actionable mutation was identified, (2) a mutation (or the absence of it) had a prognostic and/or theranostic value, and/or (3) a mutation allowing nosological precision was identified.

The main data collected were as follows: pathology, type of panel, date of each step of the process, category of prescription (diagnostic, prognostic, theranostic, or combinations thereof), and then the category of clinical decision taken following analysis of the NGS test. Only the gender and age of the patients were collected, without any other personal characteristics.

Participating NGS platforms were requested to provide information on all NGS abnormalities considered to be significant in all targeted genes/hot-spots, disease category, prescription request, and/or protocol agreements regarding actionable targets. They were also asked to transmit data on all clinical prescriptions, whether from within the hospital group or following external prescription. Practical difficulties in obtaining this information, particularly for external prescriptions, led to under-reporting of this category, with 4 laboratories only reporting internal prescriptions.

An online questionnaire to capture NGS prescription data was implemented using the clinical data management system CleanWeb (Telemedicine Technologies, Boulogne-Billancourt, France). NGS prescription data were entered by site investigators, predominantly the molecular diagnostic staff, between October 2018 and November 2020. The origin of the data was the NGS prescription itself, minutes of MDT meetings or hospital information systems. The data are stored by Telemedicine on a secure internet hosting platform conforming to ISO27001 norms. Data analysis on the anonymous data was carried out using the open-source integrated software environment R, and Microsoft Office Professional Plus Excel version 16. The study was approved by the Assistance Publique-Hôpitaux de Paris (AP-HP) ethics committee CERAPHP.5 and the promoter (AP-HP) signed a commitment of compliance with the MR003 reference methodology of the French data protection agency (CNIL) on November 30, 2017.

Cost study

A microcosting study was conducted between 2018 and 2022 at 4 molecular diagnostic platforms in France, where NGS is performed for 5 classes of hematological malignancies. The study was conducted from the health care provider perspective.^5.

After identifying types and volumes of resources required for each step of the analytic process, health economists conducted observations in situ. Data-collection spreadsheets were designed based on the workflow provided by the platforms. Personnel time, equipment and consumables (disposables and reagents), and corresponding unit costs were collected to generate an overall average cost. The calculation of production costs was for the full process up to delivering the results to the clinician with the exception of the preanalytic stage of sample collection and DNA extraction. The cost, including human resources, of archiving the complete NGS data set (eg, FASTq files) for 10 years and algorithmic pipeline filtering of data was estimated at 2 centers (No 1 and 2), including by the central bioinformatic (MOABI) platform for the 8 AP-HP laboratories. For 2 platforms (No 6 and 19), the human resources required for data archiving and pipeline filtering were included in the total cost of NGS and were not individualized.

The estimate of personnel costs was based on the time spent on each task and valued using the average AP-HP salary costs for each category of staff, based on the French annual contractual working time of 1607 hours per year. Consumable costs were based on actual consumption and real unit prices, which include negotiated reductions of catalogue prices. Following standard practice, investment costs of the materials were depreciated over 7 years (for large apparatus) or 2 years (for small apparatus)^6. using a 2.5% discount rate to reflect opportunity costs of equipment investment.^5. Hardware maintenance was calculated as 8% of the unit price of purchase. While no reruns due to problems with sampling, DNA quality, or unexpected results were noted during in situ observations, the possibility of errors requiring reruns for a specific sample was accounted for in the base case by applying a 5% markup to human resource and consumable costs to account for errors based on the expert advice from the study investigators. A deterministic sensitivity analysis was carried out to evaluate costs for a 1% and 10% markup. When it was not possible to observe a step in the process, such as analysis of the results in MDT meetings, the medical biologists were asked to estimate the time involved in such data interpretation.

Average total costs were estimated as a function of annual sample throughput between 100 and 2000 samples for a molecular diagnostic platform and were inflated by 20% to take into consideration overheads such as power, telephone, cleaning, water, logistics, and hospital administration. The estimates included 2 options regarding the allocation of equipment. While the first option assumed a theoretic point of view that the equipment was dedicated to the pathology and panel of genes being analyzed, the second one simulated a polyvalent laboratory in which the equipment was shared between different pathologies, panels, or services. For comparison between panels, an annual throughput of 500 samples for each type of NGS analysis was used for the base case analysis of costs. We estimated costs for routine practice and for complex cases as dictated by the national protocols.^7. The unit costs collected for materials in 2018 were inflated using the French consumption of care and medical goods annual indices to 2021€ values.^8. The cost analyses were performed with Microsoft Office Professional Plus Excel version 16.

Activity and impact analysis

Twenty-seven laboratories provided prospective recording of all clinical prescriptions, in the 5 categories for which they performed targeted NGS. Details of 4356 NGS prescriptions between October 2018 and November 2020 were collected, but 395 prescriptions were excluded because they were doublets (n = 10), did not respect inclusion criteria regarding pathology or date (n = 18), had insufficient annotations (n = 1), or were nonindex (ie, follow-up samples, n = 366). The latter (1–4 samples/patient), including 132 MDS or MPN, 127 acute leukemia, and 107 LPD prescriptions, were not analyzed further. The 3961 index prescriptions from 3961 patients came from 26 laboratories/hospitals, corresponding to 19 INCa platforms, because 8 belonged to the single AP-HP platform. All but 1 of the 12 mainland French regions were represented, with 1–3 platforms per region. Thus, 19 of the 28 INCa molecular diagnostic platforms were included in this study.^9.

Overall, 2584 of 3961 (65.2%) prescriptions were for myeloid disorders and 34.8% for, mainly mature B, lymphoid disorders (Table 1). Prescriptions were mainly at diagnosis/first line of treatment (69%) and showed a male predominance (58%). Age and gender by pathology (Suppl. Figure S1A-S1D) showed that 34% of the prescriptions were from patients aged 70 years or more and only 5% under 20 years. The vast majority of samples were from peripheral blood (PB) or bone marrow (BM) (96%), while lymph node or tissue-derived NGS was rare (3%), in contrast to solid tumor diagnostics. Cell-free DNA samples were not included, because they were mainly experimental in 2018–2020. NGS prescribers were essentially clinical hematologists (89%), with only 6% of prescriptions described as coming via laboratory hematologists, hematopathologists or MDTs, and 4% from other medical specialties.

After exclusion of 871 prescriptions from the 4 platforms for whom reported prescription was restricted to local samples, details of the geographic localization of prescriptions, relative to the INCa molecular platform, are illustrated in Figure 1 for the 15 platforms (3090 results) with exhaustive reporting. Of the 3090 NGS analyses, 61% were performed for patients from the same hospital/INCa platform (n = 1874) and 24% from another hospital within the same department or region (n = 745). Only 15% of samples were transmitted from another region (n = 471), mainly to Parisian and regional reference centers. Approximately half (8/15) of INCa platforms with full description of clinical NGS activity reported prescriptions from another region.

An NGS result was generated for 3945 of 3961 prescriptions, corresponding to a failure rate of 0.4%, primarily due to insufficient infiltration or failed extraction/DNA quality, mainly from LPDs (12/16 samples) and non-PB/BM samples (9/157, 5.7% compared with 7/3799, 0.18% of PB/BM; P < 0.00001). Sampling was performed on (n = 3227) or after (n = 115; 50% within 1 week) the day of NGS prescription for 87.6% of the 3813 liquid samples with available data, compared with 21.4% (27/126) of tissue samples, in keeping with a case by case approach by hematopathologists for lymphoma NGS prescription, compared with more frequent up-front NGS in liquid tumor diagnostics. For pre-existing liquid and tissue samples (576, including 6 of undefined origin), the median delay between sampling and NGS prescription was 16 days (range, 1–353), with no significant difference between liquid (median 14 days) and tissue samples (median 16 days). Results were validated and transmitted to the clinicians at a median of 10 days after performing NGS (range, 1–126 days; mean 13 days).

Detection of at least 1 clinically significant mutation was reported for 60% of the 3945 NGS results transmitted. The highest proportion was identified in AML (85%), interface MPN/MDS (82%), and ALL (77%) (Figure 2A) and the lowest in mature LPDs (43%) and MPN (47%).

Details of the impact of NGS are shown in Figure 3A and 3B. NGS platforms first classified pretest prescribing indications as follows: diagnostic, prognostic, theranostic, or combinations thereof, for 81% of samples, with the remaining 19% also integrating notions of remission monitoring, evolution, response to treatment and relapse (data not shown). NGS results were considered to impact patient management for 73.4% of the prescriptions (Figure 3A), most commonly following MDT discussions and/or protocol-driven treatment modification (Figure 3B). Impact was subclassified as useful for evaluation of prognostic risk in 34.9% (n = 1383) and necessary for treatment adaptation (actionable) in 19.6% (n = 777), but having no immediate individual therapeutic impact in 18.9% (n = 749), with details missing for 1.7% (n = 69) (Figure 3A). The distribution of perceived impact varied between disease subtypes (Figure 2B), the absence of impact being lowest in NHL/LPDs and MPN/MDS. As such, the incidence of detected mutations does not correlate with the perceived impact on patient management, as seen in LNH/LPDs. Treatment advice was specified for 1330 (33.9%) prescriptions and depended on the clinical situation, as detailed in Figure 3B. Comparison of the perceived expected (preanalysis) and actual (postanalysis) impact of NGS for the 2421 prescriptions with available information showed no-change in 59% of analyses and at most only minor modifications in 96% (Suppl. Table S2).

Cost study

Costing analyses for NGS were performed in 4 independent INCA platforms (N° 1, 2, 6, 19). NGS approaches were all targeted custom panels (amplicon or capture) and included the following: a limited first-line mature lymphoid amplicon approach (LPD/CLL N°2); a pan-hemato-oncology (Hem-Onc) capture approach (N° 19); a single disease, highly protocol-driven, approach typical for a reference center (T-ALL N°2), and 4 myeloid panel approaches, with Platform N°1 comparing manual and automated capture approaches. Unit costs as a function of annual activity are shown in Figure 4 and NGS panel details and breakdown and comparison of costs for an annual activity of 500 tests per year are shown in Table 2.

Comparison of overall costs per result showed that unit costs were higher when using dedicated equipment, particularly for lower annual throughputs (Figure 4A). With shared equipment (occupation estimated at 20%) and uniform personnel costs (AP-HP rates), unit costs at lower levels of activity were dependent on the annual throughput for each type of NGS panel.

The average overall cost per sample was, as expected, much lower for the restricted, first-line mature lymphoid LPD/CLL amplicon panel (191€; 5% rerun rate) compared with all others (Table 2), when the cost of capture panels varied from 368€ to 513€. Unit costs varied more as a function of the number of targets reported (5–102) than kb (24–780) sequenced. Unit costs varied from 0.5 to 7.9€ per kb sequenced, from 3.6€ to 14.7€ per target gene/hot-spot sequenced, and from 4.3 to 73.8€ per sample for target gene/hot-spot reported. Extremes are represented by the 2 formats used to report data from the T-ALL panel. Reporting of only the 5 actionable T-ALL genes, compared with reporting relevant mutations for all 102 genes in relapse/resistant or complex cases (Figure 4B) led to a doubling of human resource unit costs (essentially medical) from 63€ to 124€ and an increase in overall costs of 20% (369€ to 443€). Based on the local estimation that 75% of samples require reporting on only 5 targets and 25% require full analysis, the average unit cost would be 387€ per sample. This was similar to the unit cost of 393€ using the pan-Hem-Onc panel (platform N° 19), reporting on up to 44 genes, with the actual number reported depending on the prescription and clinical situation. Comparison of manual versus automated costs for a myeloid capture panel on platform N°1 (Figure 4C) showed that overall unit costs were slightly higher following automation at higher annual throughput, with a significant increase in the cost of consumables, partially offset by a decrease in human resources, which can therefore be diverted elsewhere, and equipment costs. This was, however, reversed at lower levels of activity. This can be explained by the low number of samples processed in a single run, which results in higher sequencing costs but makes it possible to obtain a result within 48 hours of receiving the sample. Automation also allows for more robust and reproducible NGS workflows.

The cost of data storage (eg, FASTQ formatted files) and algorithmic pipeline filtering before validation was not the focus of this cost study but was estimated by centers 1 and 2 and is shown in Table 2 and Suppl. Table S3. Centers 6 and 19 included these costs in the overall cost estimate for the NGS analysis.

The activity thresholds for which a change in ±50 annual NGS analyses did not result in a change in unit costs of >±1% (flat part of the cost/activity curve) ranged between 550 and 700 analyses per year, except for the automated myeloid panel (450/year.).

Taken together, these costing analyses demonstrate that minimal annual activity levels of over 500 samples are acceptable from an economic point of view, but that automation may allow lower activity levels while maintaining resource efficiency. They also show that a single, uniform reimbursement policy for targeted NGS is not appropriate, because it is unlikely to incite economic optimization.

National consensus for NGS prescribing

The GBMHM published consensus NGS panels for lymphoid cancers in 2019.^17. At the request of the HAS, these were complemented in 2022 by recommendations on indications for analysis in routine clinical practice, after concertation with the relevant national clinical cooperative groups, under the aegis of the SFH. These recommendations included NGS indications for immunogenetics (IG/TR) and cell-free/circulating DNA, neither of which are addressed here.

Hematological cancers were divided into 20 categories (Table 3), including 11 lymphoid and 9 myeloid, with an annual incidence ranging from 100 (mastocytosis with hematological involvement) to 5442 (myeloma).^7. With over 40,000 new cases of hematological cancer reported each year in France (population 67 million inhabitants in 2018/2019), NGS was recommended for at least 10% of cases in virtually all disease categories, but also for identification of prognostic markers at diagnosis in all cases of CLL, ALL, AML, MDS, MPN/MDS, and myelofibrosis. The latter group corresponded to a total of 15,796 new cases per year. Multiple myeloma was analyzed in a single national reference center, essentially within clinical trials, and, as such, was not detailed, despite its frequency, but was here considered to require NGS in ≈10% of cases at diagnosis. Combining these indications, it would be expected that at least 18,300 NGS oncogenetic analyses (15,800 plus 10% of the remaining 24,800) would be prescribed each year in France for hematological cancers. This is an underestimate of the total NGS activity reported each year, because it did not take into account analysis at (suspected) therapeutic resistance/evolution/relapse, measurable/minimal residual disease monitoring or IG/TR analyses.

In keeping with the cost analyses and types of targeted panels developed by diagnostic laboratories, recommendations for panel use have been divided into panels targeting ≤20 or more genes/hot-spots (Table 3). Individual panel use is up to local choice for custom in-house IVD or CE-IVD NGS assays, in turn dependent on local first-line versus reference recruitment.

These national consensus panels, approved by both the clinical and diagnostic hematology-oncology community, should contribute to the identification of appropriate prescribing practice of targeted NGS and its reimbursement.