Cost-Effectiveness of Pharmacogenomics-Guided Prescribing to Prevent Gene-Drug-Related Deaths: A Decision-Analytic Model

Aim: Prospective studies support the clinical impact of pharmacogenomics (PGx)-guided prescribing to reduce severe and potentially fatal adverse effects. Drug-gene interactions (DGIs) preventing potential drug-related deaths have been categorized as “essential” by the Dutch Pharmacogenetics Working Group (DPWG). The collective clinical impact and cost-effectiveness of this sub-set is yet undetermined. Therefore, we aim to assess impact and cost-effectiveness of “essential” PGx tests for prevention of gene-drug-related deaths, when adopted nation-wide. Methods: We used a decision-analytic model to quantify the number and cost per gene-drug-related death prevented, from a 1-year Dutch healthcare perspective. The modelled intervention is a single gene PGx-test for CYP2C19, DPYD, TPMT or UGT1A1 to guide prescribing based on the DPWG recommendations among patients in the Netherlands initiating interacting drugs (clopidogrel, capecitabine, systemic fluorouracil, azathioprine, mercaptopurine, tioguanine or irinotecan). Results: For 148,128 patients initiating one of seven drugs in a given year, costs for PGx-testing, interpretation, and drugs would increase by €21.4 million. Of these drug initiators, 35,762 (24.1%) would require an alternative dose or drug. PGx-guided prescribing would relatively reduce gene-drug related mortality by 10.6% (range per DGI: 8.1–14.5%) and prevent 419 (0.3% of initiators) deaths a year. Cost-effectiveness is estimated at €51,000 per prevented gene-drug-related death (range per DGI: €-752,000–€633,000). Conclusion: Adoption of PGx-guided prescribing for “essential” DGIs potentially saves the lives of 0.3% of drug initiators, at reasonable costs.


Supplementary Table 2 Systematic methodology to select publications and extract absolute risk of gene-drug-related-death
The steps shown in Table 1 are performed systematically to select relevant publications from which to extract the absolute risk of gene-drug-related death. Risk extraction is performed by using methodology corresponding to that step. Each extracted absolute risk of death is given a certainty score based on the step in which publication(s) are selected.
The publication selection is performed systematically using only the publications listed in the summary of the systematic review of literature underlying the DPWG guideline ("the risk analysis"). Each of the publications listed in the risk analysis have been scored systematically by the DPWG both on the clinical relevance and on the quality of evidence [1]. The quality of evidence for each publication was scored on a five-point scale ranging from 0 (lowest quality of evidence) to 4 (highest quality of evidence). Score 4 corresponds to controlled, published studies of good quality or well-performed meta-analyses. Good quality is defined as: it is known whether comedication with an influence on the phenotype has been used; it is known whether other confounders are present (depending on the substance, for example smoking or not); the data are based on steady state kinetics; corrected for this at a variable dose [2]. Score 3 corresponds to controlled, published studies of moderate quality or poorly performed meta-analyses (for example, no good statistics, studies with different measured endpoints, heterogeneity, publication bias). Moderate quality is defined as: at least one of the criteria considered under good quality does not apply [2].
The risk of gene-drug-related death will vary across predicted phenotype groups. For example, risk of fluoropyrimidine-induced toxicity increases with decreasing DPYD gene activity scores (GAS), when all groups receive the same initial dose. Furthermore, when a PGx test is used to guide dose selection, those who have an actionable predicted phenotype (DPYD GAS 0-1.5) will have a reduced risk of fluoropyrimidine-induced toxicity when compared to risk when using a normal dose. The risk of death as a result of fluoropyrimidine-induced toxicity, however, in those with a non-actionable predicted phenotype (in this case DPYD GAS 2) will have the same risk, regardless of being PGx tested. Therefore, we will extract the absolute risk of death for each predicted phenotype category, across three groups: 1) tested-actionables (e.g. DPYD GAS 0, 0.5, 1 and 1.5 with PGx informed reduced dose), 2) non-actionables (e.g. DPYD GAS 2 with normal dose) and 3) untested-actionables (e.g. DPYD GAS 0, 0.5, 1 and 1.5 with normal dose). The predicted phenotype-gene interactions which are categorized as being actionable or non-actionable are provided in Table 1.
Other publications may be selected for extraction of each absolute risk. For example, risks of untestedactionables and non-actionables groups may be extracted from observational studies. However, the risks of tested-actionables group must be extracted from interventional studies. When a publication is selected for one of these three groups within one step, but is not suitable for risk extraction of the remaining groups, the following step is performed to find a suitable publication for the remaining groups. Table 1: Systematic methodology to select suitable publications and subsequent extraction of absolute risk of gene-drug-related-death within one year. The steps are executed consecutively until at least one suitable publication is found. Step

Suitable publication(s) Risk extraction method 1) tested actionables 2) nonactionables 3) untested actionables
Certainty Score 1 Publications reporting predicted phenotype group: quality score 4 a , powered on mortality The risk of mortality of the most severe preventable clinical consequence within one year is extracted.

= Very certain 2
Publications reporting predicted phenotype group: quality score 4 a The risk of the intermediary outcome within one year is extracted and is multiplied by the risk of death as a result of this intermediary outcome within one year. This is found by searching literature.

= Certain 3
Publications reporting predicted phenotype group: quality score 3 b , powered on mortality The risk of mortality of the most severe preventable clinical consequence within one year is extracted.

= Fairly certain 4
Publications reporting predicted phenotype group: quality score 3 b The risk of the intermediary outcome within one year is extracted and is multiplied by the risk of death as a result of this intermediary outcome within one year. This is found by searching literature.

= Uncertain 5
Perform literature review in review of a usable study regarding the relevant DGI When the study is powered on mortality the risk of mortality within one year is extracted. When the study reported on an intermediary outcome, the intermediary outcome within one year is extracted and is multiplied by the risk of death as a result of this intermediary outcome within one year. This is found by searching literature.
Based on quality score criteria of DPWG 6 No publication selected Estimation 0 = Very uncertain a Controlled, published studies of good quality with genotyping and / or phenotyping in patients / healthy subjects with clinical endpoints (effectiveness, side effects) or relevant kinetic endpoints (change in plasma level, AUC, half-life, etc.) or good performed meta-analyzes. Good quality is defined as: it is known whether comedication with an influence on the phenotype has been used; it is known whether other confounders are present (depending on the substance, for example smoking or not); the data are based on steady state kinetics; corrected for this at a variable dose [2]. b Controlled, published studies of moderate quality. with genotyping and / or phenotyping in patients / healthy subjects with clinical endpoints (effectiveness, side effects) or relevant kinetic endpoints (change in plasma level, AUC, half-life, etc.) or poor performed meta-analyzes (for example, no good statistics, studies with different measured endpoints, heterogeneity, publication bias). Moderate quality means that one or more of the items considered under good quality are missing [2].

Publication(s) selection
Publications are selected only if they present usable risk data and are sufficiently representative for the healthcare system and patients in the Netherlands. Being usable is defined as presenting risk data from which risks for at least one of the three groups can be calculated without requesting raw data underlying the publication. Being sufficiently representative is defined as studies including patients of which at least 50% are from North America or Europe.

Absolute risk extraction
Once at least one publication has been selected for each relevant drug-phenotype category for three patient groups: 1) tested-actionables, 2) non-actionables and 3) untested-actionables we are able to extract risks. This is performed corresponding to the step in which the publication was selected (see below). Within a particular step, if only one publication is selected, the absolute risks of death are extracted from that single publication. When more than one publications are found suitable, the absolute risks of death are extracted from each publication and the mean is taken (weighed by the number of patients). However, when multiple meta-analyses are selected within one step, the risk extraction will only be performed based on the most recent meta-analysis, provided the majority of studies included in older meta-analyses.
Step 1: Publications reporting predicted phenotype group: quality score 4, powered on mortality (certainty score 4) The risk of mortality of the most severe preventable clinical consequence within one year is extracted directly.
Step 2: Publications reporting predicted phenotype group: quality score 4, calculating the risk of death from intermediary outcome (certainty score 3) The risk of an intermediary outcome within one year is extracted and is multiplied by the risk of death as a result of this intermediary outcome within one year. Risk of death as a result of an intermediary outcome is found by searching literature and presented in Appendix 2 section "Assessment of risk of drug-related death following an intermediary outcome associated with the gene-drug interaction".
Step 3: Publications reporting predicted phenotype group: quality score 3, powered on mortality (certainty score 2) The risk of mortality of the most severe preventable clinical consequence within one year is extracted.
Step 4: Publications reporting predicted phenotype group: quality score 3, calculating the risk of death from intermediary outcome (certainty score 1) The risk of the intermediary outcome within one year is extracted and is multiplied by the risk of death as a result of this intermediary outcome within one year. Risk of death as a result of an intermediary outcome is found by searching literature and presented in Appendix 2 section "Assessment of risk of drug-related death following an intermediary outcome associated with the gene-drug interaction".

Step 5: Perform literature review in review of a usable study regarding the relevant DGI (certainty score is based on quality of evidence criteria of DPWG)
When the study is powered on mortality the risk of mortality within one year is extracted. When the study reported on an intermediary outcome, the intermediary outcome within one year is extracted and is multiplied by the risk of death as a result of this intermediary outcome within one year. This is found by searching literature and presented in Appendix 2 section "Assessment of risk of drug-related death following an intermediary outcome associated with the gene-drug interaction".
Step 6: No publication selected: (certainty score 0 -estimation) When none of the selected publications are intervention studies, we are unable to extract the risk of death for tested actionables. In this case we estimate the risk of death for tested actionables to equal the risk of death of non-actionables. In this case it is given a certainty score of 0 (estimation).

References
Conclusion of selected publications and absolute risks extracted (death as a result of diarrhoea): 45 1.5.7 Conclusion of selected publications and absolute risks extracted (sum absolute risk of death due to neutropenia and absolute risk of death due to diarrhoea Assessment of risk of drug-related death following an intermediary outcome associated with the gene-drug interaction . Since the risk analysis is combined for both azathioprine and mercaptopurine, the publication selection and risk extraction will also be combined for both.
There is a DPWG guideline for the indications of acute lymphoblastic leukaemia (ALL) and irritable bowel syndrome (IBD

Absolute risk extraction non-actionables (EM):
Booth RA et al. Risk of leukopenia was 0.209573847 (See Appendix Figure 2, sum of events/sum of patients = 359/1713) among non-actionable TPMT EMs. Risk of death among IBD patients who develop myelotoxicity is approximately 0.01 (37). Therefore, risk of death as a result of leukopenia is 0.209573847 x 0.01 = 0.002095738 for non-actionable TPMT EMs. These are given a certainty score of 3.

Absolute risk extraction untested actionables (IM and PM):
Booth RA et al.   (37). Therefore, risk of death as a result of leukopenia is 1 x 0.01 = 0.01 for untested TPMT PMs. These are given a certainty score of 3.
In the genotype-guided group, EMs received the normal thiopurine dose and IMs 50% of the normal dose. PM were scheduled to receive 0-10% of the normal dose. Hematologic adverse events were defined as leukocyte count < 3.0x109/L or platelet count < 100x109/L.A significantly smaller proportion of carriers of the TPMT variants in the intervention group (2.6%) developed hematologic ADRs compared with patients in the control group (22.9%) (relative risk, 0.11; 95% confidence interval, 0.01-0.85). TPMT IM and PM: Coenen et al. has combined the TPMT IMs and PMs in one group, therefore we will also perform risk extraction for IM and PMs combined. Risk of hematologic adverse events was 0.025641026 among tested TPMT IMs and PMs (1 event among 39 patients, see Table 3). Risk of death among IBD patients who develop myelotoxicity is approximately 0.01 (37). Therefore, risk of death as a result of leukopenia is 0.025641026 x 0.01 = 0.025641026 for tested TPMT IMs and PMs. These are given a certainty score of 1. There are no studies available through the "risk analysis" that have a quality score of 4.

Conclusion:
No literature was selected therefore we will continue to the next step.
2 Publications reporting predicted phenotype group: quality score 4 There are no studies available through the "risk analysis" that have a quality score of 4.

Conclusion:
No literature was selected therefore we will continue to the next step. We found no additional studies through our own literature search. Therefore, we estimated the absolute risk on death for thioguanine to be similar to azathioprine and 6mercaptopurine. The certainty score given is 0, since it is an estimation. Since the risk analysis is combined for both capecitabine and 5-FU, the publication selection and risk extraction will also be combined for both.

Conclusion of selected publications and absolute risks extracted:
Steps performed systematically to select suitable publication(s) form which extraction is performed Publications reporting predicted phenotype group: quality score 4, powered on mortality Conclusion: No publication was selected therefore we will continue to the next step.

Study Conclusion
Deenen MJ et al. (42) Not powered on mortality. Not selected.
Meulendijks D et al.
Not powered on mortality. Not selected.
Meulendijks D et al. (44) Not powered on mortality. Not selected.
Rosmarin D et al. (45) Not powered on mortality. Not selected.
Terrazzino S et al.
Not powered on mortality. Not selected.
Vulsteke C et al. (47) Not powered on mortality. Not selected.
2 Publications reporting predicted phenotype group: quality score 4

Study Conclusion
Deenen MJ et al.
Not powered on mortality. Not selected.
van Kuilenburg AB et al. (74) Not powered on mortality. Not selected.

Conclusion:
No publications were selected therefore we will continue to the next step. 4 Publications reporting predicted phenotype group: quality score 3
Usable risk data: Yes Selected for extraction of tested groups.
Henricks LM et al.

Absolute risk extraction non-actionables (GAS 2):
Meulendijks D et al. (44) was selected for extraction of non-actionables groups. Risk of grade 3 or higher fluoropyrimidine induced toxicity was unable to be extracted for untestedactionable DPYD GAS 0.5 from Meulendijks D et al. (44). No suitable publication was identity in steps 3 or 4. Therefore we will assume the risk of grade 3 or higher fluoropyrimidine induced toxicity to increase linearly with decreasing GAS. Delta risk of death between GAS 1.5 and GAS 1.0 was 0.005176056 -0.003377863 = 0.0018. Therefore we estimate the risk of grade 3 or higher fluoropyrimidine induced toxicity for GAS 0.5 to be 0.005176056 -0.0018 = 0.0034. Therefore we estimate the risk of grade 3 or higher fluoropyrimidine induced toxicity for GAS 0.5 to be [risk of death GAS 1.5 + delta risk] = 0.005176056 + 0.0018 = 0.0070. These are given a certainty score of 0 GAS 0 (*2A/*2A or *13/*13 or *2A/*13): Risk of grade 3 or higher fluoropyrimidine induced toxicity was unable to be extracted for untestedactionable DPYD GAS 0. from Meulendijks D et al. (44). No suitable publication was identified in steps 3 or 4. Therefore we will assume the risk of grade 3 or higher fluoropyrimidine induced toxicity to increase linearly with decreasing GAS. Delta risk of death between GAS 1.5 and GAS 1.0 was 0.005176056 -0.003377863 = 0.0018. Therefore we estimate the risk of grade 3 or higher fluoropyrimidine induced toxicity for GAS 0.5 to be [risk of death GAS 0.5 + delta risk] = 0.0070 + 0.0018 = 0.0088. These are given a certainty score of 0.

Absolute risk extraction tested actionables (GAS 0-1.5):
Kleinjan JP et al. (48) Risk of death as a result of grade 3 fluoropyrimidine induced toxicity is approximately 0.0075 (75). Therefore, risk of death as a result of leukopenia is 0.131868 x 0.0075 = 0.0010 for untestedactionable DPYD GAS 1.5. These are given a certainty score of 1.  Risk of grade 3 or higher fluoropyrimidine induced toxicity was unable to be extracted for untestedactionable DPYD GAS 0.5 from (48) (49, 50) (51,55). No suitable publication was identified in step 5. Therefore we will assume the risk of grade 3 or higher fluoropyrimidine induced toxicity is equal to the mean risk of death of GAS 1.5 and 1. The mean of these is 0.0012. These are given a certainty score of 0. Risk of grade 3 or higher fluoropyrimidine induced toxicity was unable to be extracted for untestedactionable DPYD GAS 0.5 from (48) (49, 50) (51,55). No suitable publication was identified in step 5. Therefore we will assume the risk of grade 3 or higher fluoropyrimidine induced toxicity is equal to the mean risk of death of GAS 1.5 and 1. The mean of these is 0.0012. These are given a certainty score of 0.

Conclusion of selected publications and absolute risks extracted:
Not powered on mortality. Not selected.
Shuldiner AR et al.
Not powered on mortality. Not selected.
Frére C et al. (110) Not powered on mortality. Not selected.
Aleil B et al. (111) Not powered on mortality. Not selected.
Sibbing D et al. (112) Not powered on mortality. Not selected.
Brackbill ML et al.
Not powered on mortality. Not selected.
Giusti B et al. (114) Not powered on mortality. Not selected.
Umemura K et al.
Not powered on mortality. Not selected.
Frére C et al. (116) Not powered on mortality. Not selected.

Conclusion:
No publication was selected therefore we will continue to the next step.

Absolute risk extraction non-actionables (UM and EM):
Sorich MJ et al.  (79) has combined the CYP2C19 IMs and PMs in one group, therefore we will also perform risk extraction for IM and PMs combined. Risk of major adverse cardiovascular outcomes was 0.107436901 (See Figure 2, white non-PCI + white PCI, sum of events/sum of patients = 177+619/1891+5518) among untested actionable CYP2C19 IMs and PMs. The risk of cardiovascular death in MACE is 0.34 (120). Therefore, risk of death as a result of adverse cardiovascular events is 0.107436901 x 0.34 = 0.036807086 for untested actionable CYP2C19 IMs and PMs. These are given a certainty score of 3.

Absolute risk extraction tested actionables (IM and PM):
Lee CR et al. (97) Figure 3A, LOF-alt, n events/n patients (extracted from Fig 1A)  The risk of cardiovascular death in MACE is 0.34 (120). Therefore, risk of death as a result of adverse cardiovascular events is 0.071428571x 0.34 = 0.024470899 for tested actionable CYP2C19 IMs and PMs. These are given a certainty score of 1.

Conclusion of selected publications and absolute risks extracted:
Actionability We found no intervention studies through our own literature search. Therefore, we estimated the absolute risk on death to be equal to that of non-actionables. These are given a certainty score of 0.
Neutropenia: Risk of grade 3 or higher neutropenia was 0.1121 (See Figure 2, b, high IRI, sum of events/sum of patients = 72/642) among non-actionable *1/*1. Risk of drug-related death as a result of myelosuppression is 0.00949 (section treatment related deaths: 1.3% died of treatment related effects, of which 73% was associated with myelosuppression) (162). Therefore, risk of death as a result of grade 3 neutropenia is 0.1121 x .,00949 = 0.001064299 for non-actionable *1/*1. These are given a certainty score of 3.
Risk of grade 3 or higher neutropenia was 0.1865 (See Figure 3, b, high IRI, sum of events/sum of patients = 102/547) among non-actionable *1/*28 and IMs. Risk of death as a result of myelosupression is 0.00949 (section treatment related deaths: 1.3% died of treatment related effects, of which 73% was associated with myelosuppression) (162). Therefore, risk of death as a result of grade 3 neutropenia is 0.1865 x 0.00949 = 0.001769616 for non-actionable *1/*28 and IM. These are given a certainty score of 3. Diarrhoea: Risk of grade 3 or higher diarrhoea was 0.1109 (See Figure 4, b, high IRI, sum of events/sum of patients = 73/658) among non-actionable *1/*1. Risk of drug-related death as a result of diarrhoea is 0.001363473 (section treatment related deaths sum of patients death of diarrhoea/total patients = 19/13935) . Therefore, risk of death as a result of grade 3 diarrhoea is 0.1109 x 0.0013 = 0.000151267 for non-actionable *1/*1. These are given a certainty score of 3.
Risk of grade 3 or higher diarrhoea was 0.1473 (See Figure 5, b, high IRI, sum of events/sum of patients = 80/543) among non-actionable *1/*28 and IM. Risk of drug-related death as a result of diarrhoea is 0.001363473 (section treatment related deaths sum of patients death of diarrhoea/total patients = 19/13935) (162). Therefore, risk of death as a result of grade 3 diarrhoea is 0.1473 x 0.0013 = 0.00020088 for non-actionable *1/*28 and IM. These are given a certainty score of 3.

Absolute risk extraction untested actionables (*28/*28 and PM):
Liu X et al. (125). is a meta-analysis of 16 studies including a total of 2,328 mainly Caucasian patients with colorectal cancer. The outcome measure was grade 3-4 toxicity. Neutropenia: Risk of grade 3 or higher neutropenia was 0.3525 (See Figure 2, b, high IRI, sum of events/sum of patients = 43/122) among untested *28/*28 and PM. Risk of drug-related death as a result of myelosupression is 0.00949 (section treatment related deaths: 1.3% died of treatment related effects, of which 73% was associated with myelosuppression) (162). Therefore, risk of death as a result of grade 3 neutropenia is 0.3525 x 0.00949 = 0.003344836 for untested *28/*28 and PM. These are given a certainty score of 3.

Diarrhoea:
Risk of grade 3 or higher diarrhoea was 0.2155 (See Figure 4, b, high IRI, sum of events/sum of patients = 25/116) among untested *28/*28 and PM. Risk of drug-related death as a result of grade 3 diarrhoea is approximately 0.001363473 (section treatment related deaths sum of patients death of diarrhoea/total patients = 19/13935) (162). Therefore, risk of death as a result of grade 3 diarrhoea is 0.2155 x 0.001 = 0.000293852 for untested *28/*28 and PM. These are given a certainty score of 3.

Absolute risk extraction tested actionables (*28/*28 and PM):
Since no intervention studies were identified, we estimate the risk of death for tested-actionables to equal the risk of death of non-actionables (*1/*1). In this case it is given a certainty score of 0 (estimation).