A systematic review and meta‐analysis of the impacts of germline pharmacogenomics on severe toxicity and symptom burden in adult patients with cancer

Abstract

The clinical application of Pharmacogenomics (PGx) has improved patient safety. However, comprehensive PGx testing has not been widely adopted in clinical practice, and significant opportunities exist to further optimize PGx in cancer care. This systematic review and meta‐analysis aim to evaluate the safety outcomes of reported PGx‐guided strategies (Analysis 1) and identify well‐studied emerging pharmacogenomic variants that predict severe toxicity and symptom burden (Analysis 2) in patients with cancer. We searched MEDLINE, EMBASE, CENTRAL, clinicaltrials.gov, and International Clinical Trials Registry Platform from inception to January 2023 for clinical trials or comparative studies evaluating PGx strategies or unconfirmed pharmacogenomic variants. The primary outcomes were severe adverse events (SAE; ≥ grade 3) or symptom burden with pain and vomiting as defined by trial protocols and assessed by trial investigators. We calculated pooled overall relative risk (RR) and 95% confidence interval (95%CI) using random effects models. PROSPERO, registration number CRD42023421277. Of 6811 records screened, six studies were included for Analysis 1, 55 studies for Analysis 2. Meta‐analysis 1 (five trials, 1892 participants) showed a lower absolute incidence of SAEs with PGx‐guided strategies compared to usual therapy, 16.1% versus 34.0% (RR = 0.72, 95%CI 0.57–0.91, p = 0.006, I ² = 34%). Meta‐analyses 2 identified nine medicine(class)‐variant pairs of interest across the TYMS, ABCB1, UGT1A1, HLA‐DRB1, and OPRM1 genes. Application of PGx significantly reduced rates of SAEs in patients with cancer. Emergent medicine‐variant pairs herald further research into the expansion and optimization of PGx to improve systemic anti‐cancer and supportive care medicine safety and efficacy.

INTRODUCTION

Clinical pharmacogenomics (PGx) plays an increasingly important role in patient safety. There is abundant evidence demonstrating that comprehensive germline PGx‐guided medicine optimization strategies reduce severe adverse events (SAEs) in general medicine patients. ¹ , ² Such strategies, supported by guidelines such as those published by the Dutch Pharmacogenetics Working Group (DPWG) ³ or Clinical Pharmacogenetics Implementation Consortium (CPIC), ⁴ have been associated with especially high effect size in cancer care but studies were inadequately powered to show this safety benefit. ²

Patients with cancer frequently receive high‐risk medicines for the treatment of their cancer and management of associated symptoms. Relative to community averages, they are also likely to be older and have other co‐morbidities contributing to polypharmacy (prescribed five or more medicines). Therefore, this patient group has a high frequency of adverse medicine events and symptom‐driven complaints, leading to unplanned emergency room presentations and unplanned hospital admissions. ⁵ As the global burden of cancer increases, ⁶ pre‐emptive and preventative strategies such as those offered through the promise of PGx become increasingly important to reduce patient morbidity and improve the efficiency of health care resource utilization. Although established PGx strategies deliver significantly improved outcomes, ¹ , ² there remain considerable opportunities in cancer care to build on current evidence to improve the specificity and sensitivity of germline pharmacogenomic testing and reduce further the overall toxicity and symptom burden associated with anti‐cancer and supportive care medicines. At the time of this review, there were 77 medicines associated with a CPIC/DPWG recommendation, of which six are systemic anti‐cancer therapies (irinotecan, tamoxifen and classes fluoropyrimidines, and thiopurines) and another 20 are supportive care medicines frequently initiated within a cancer care setting. ⁷

The primary objectives of this meta‐analysis are to (1) evaluate in patients with cancer the safety outcomes of CPIC‐ or DPWG‐endorsed PGx‐guided strategies compared to usual therapy, and (2) synthesize the evidence for other clinically relevant pharmacogenomic variants that may predict severe toxicity and symptom burden that are not included among current CPIC or DPWG recommendations. A secondary objective is to evaluate the impact of PGx‐guided strategies on radiologically confirmed objective disease response to treatment within studies.

METHODS

Data sources and search strategy

Advanced literature searches were completed in April 2023 using the following electronic databases: Medline (Ovid), Embase (Ovid), and Cochrane CENTRAL (Wiley), as well as the clinical trials registers clinicaltrials.gov and International Clinical Trials Registry Platform (ICTRP). Articles were limited to English language only with no publication year restrictions. The search strategy was developed using the PICO framework specific to each database with oversight from a medical librarian (SLa). In Medline, the search strategy consisted of a combination of exploded subject headings (MESH) and various keywords to best identify the target literature (Table S1). Subject headings applied in Medline included: “Pharmacogenetics,” “Neoplasms,” as well as “randomized controlled trials,” “clinical trials,” and “comparative studies.” The subjects were combined in their associated cluster groups with keywords where all word variations were searched. The adjacency operator was applied in some instances that linked words in proximity. Filters were applied to ensure studies included adults. The “AND” operator was then applied to combine all separate concepts and yield the final relevant publications. The search in Embase followed a similar strategy with variations according to the subject thesaurus (Emtree). In Cochrane CENTRAL, ClinicalTrials.gov and ICTRP, keyword combinations were used. All published literature until January 31, 2023 was captured including any errata.

Data extraction and quality of assessment

All publications were downloaded, and then screened using Covidence. Non‐English abstracts and abstracts grossly out of scope were excluded. Full‐text review applying inclusion/exclusion criteria was undertaken by two independent reviewers (SL and MS). Disagreement was resolved by consensus with a third author, CK. Study selection, data checking, and reporting were undertaken according to the Preferred Reporting Items for Systematic Review and Meta‐Analyses (PRISMA). ⁸

Eligibility criteria and study selection

Two separate analyses were performed, first to evaluate the impact of PGx strategies on safety (Analysis 1) using interventional data, and second to explore evidence for a clinically relevant association between safety outcomes and other studied variants not yet associated with PGx recommendations (Analysis 2) using observational data.

1. Analysis 1: eligible studies were randomized or non‐randomized controlled trials (including cluster randomized trials and secondary analyses of clinical trials) and comparative studies for comparing PGx versus usual therapy. Only studies that conformed to CPIC or DPWG recommendations for actionable medicine‐gene pairs were included, with comparisons made between equivalent groups, either variant and wild‐type carriers or variant carriers alone.

2. Analysis 2: Where no PGx guideline is established, eligible studies compared safety outcomes for variant carriers versus wild‐type carriers. Any clinical trials were included.

Other inclusion criteria were that participants must have received active treatment with systemic anti‐cancer therapy, radiotherapy, or surgery, participants were required to be aged ≥15 years to include adolescents and adults, at least one cohort must have utilized germline pharmacogenomic testing with genotyping performed on non‐tumor tissue (such as peripheral blood or saliva), and studies must have reported at least one of the primary outcomes; severe toxicity defined as Grade ≥3 according to the Common Terminology Criteria for Adverse Events (CTCAE) definitions and grades, or as otherwise classified within individual studies, or symptom burden defined according to validated visual analog scales such as Brief Pain Inventory (BPI) or rates of complete emesis control. Where there were multiple reports from a single trial, only a single report – one with the highest number of total cases or events relevant to the primary outcome comparison was included.

Studies were excluded if there was no full text (abstracts only/conference proceedings), if they were small with a sample size <25, the mean age of patients was <15 years, or if patients received an allogeneic transplant or other donor tissue, or had no diagnosis of cancer, or if only tumor tissue was genotyped, and if outcomes relating to severe toxicity or symptom burden were not reported. PGx discovery‐only studies where putative associations were not subsequently replicated or validated prospectively were also excluded. Reports of cancer prognostic genes (e.g., BRCA1), and tumor or somatic genes (e.g., BRAF) comparisons were excluded.

Assessment of risk of bias in included studies

Risk of bias (RoB) was assessed using validated tools applicable to the design of the study. The Risk Of Bias In Non‐randomized Studies of Interventions (ROBINS‐I) tool ⁹ and revised tool to assess RoB in randomized trials (RoB 2) ¹⁰ were utilized to assess the studies considered in Analysis 1 (comparison of PGx‐guided dosing vs. usual therapy for CPIC/DPWG‐endorsed medicine‐gene pair recommendations). For Analysis 2 (comparison of variant carriers vs. wild‐type carriers) the Newcastle‐Ottawa Quality Assessment Scale (NOQAS) ¹¹ was used to assess cohort and case–control studies within clinical trials, with overall scores classified as low (8–9), moderate (7–6) or critical (≤5). NOQAS like ROBINS‐I is recommended by the Cochrane Scientific Committee for non‐randomized studies and is more relevant for non‐interventional studies. Three independent reviewers (SL, CK, JN) performed the quality assessment. Disagreement was assessed by additional reviewers (co‐authors) for consensus. Certainty assessment for findings of pooled analyses was independently conducted by two authors (SL, CK) with an additional reviewer (JN) mediating any disagreements according to the GRADE approach and summarized using GRADE Evidence Profiles. ¹² Any differences in anti‐cancer therapy intensity or dose–response gradients were considered as part of the indirectness assessments for GRADE.

Outcome measures

Two primary analyses were conducted: PGx versus Usual Therapy (Analysis 1), and variant carriers versus wild‐type carriers (Analysis 2). The two primary outcomes were the proportion of patients with severe adverse events (SAEs; Analysis 1 and Analysis 2) and symptom burden assessments (Analysis 2) occurring during study follow‐up. The proportion of study participants who demonstrated an at‐risk genotype was calculated if all studies within the subset analysis confirmed conformance with Hardy–Weinberg equilibrium. For studies that reported more events than patients, an event rate per patient years was calculated based on the duration of patient follow‐up reported for the study. Sensitivity analyses were conducted to determine any impact of this variation on calculated risk ratios. For symptom burden, brief pain inventory scores were used for pain, and rates of complete emesis control were used for vomiting. A secondary analysis was conducted for studies that did not provide adequate information for case or event rate data to be aggregated. A secondary outcome measure for Analysis 1 was the objective disease response rate based on the Response Evaluation Criteria in Solid Tumors (RECIST) Version 1.1 in studies reporting safety.

Statistical analysis

Descriptive statistics relating to outcomes of toxicity or symptom burden for variant carriers versus wild‐type carriers and outcomes of toxicity and overall survival for PGx versus Usual therapy were summarized. Continuous variables were described as mean and standard deviation; dichotomous variables were described as counts and percentages. Meta‐analyses were performed for outcome comparisons between (a) PGx versus Usual Therapy and (b) variant carriers versus wild‐type carriers. Data was pooled and analyzed using Review Manager 5.4 with random effects methods applied to ensure conservative estimates of statistical significance given the expected heterogeneity between studies. Effect measures for all outcomes were reported as Risk Ratio (RR) with 95% confidence intervals (95% CI). Heterogeneity was quantified using the homogeneity test (χ ² with k‐1 df; p = 0.1 for significance) and I ²statistic. Due to the small number of trials, and expected clinical heterogeneity based on the patient cohorts, the I ² statistic has been emphasized.

RESULTS

Studies identification and characteristics

The search strategy identified 6811 studies from EMBASE (n = 2439), Cochrane Library (n = 1508), and MEDLINE (n = 2243) (Figure 1). A total of 1676 duplicates were excluded. A further 4585 studies were excluded as irrelevant based on a review of the title and abstract. Following full‐text review, a further 438 studies were excluded due to different outcome measures (n = 164), other study design (n = 57), somatic genes or prognostic genes (n = 52), non‐full‐text (n = 32), non‐implementation studies for CPIC/DPWG‐endorsed PGx‐guided strategies (n = 26), other intervention (n = 24) and discovery only associations without validation or replication (n = 24), sample size <25 (n = 10). Two studies were excluded from Analysis 1 as these reported results from another included trial. ¹³ , ¹⁴

FIGURE 1.

PRISMA flow diagram. Preferred reporting items for systematic reviews and meta‐analyses (PRISMA) flow diagram. CPIC, clinical pharmacogenetics implementation consortium; DPWG, Dutch pharmacogenetics working group; FP, fluoropyrimidines (e.g., 5‐fluorouracil, capecitabine); 5HT3A, 5HT3‐antagonists (e.g., ondansetron, granisetron, palonosetron); MTX, methotrexate; R‐CHOP = rituximab, cyclophosphamide, doxorubicin, vincristine, prednisolone.

The research strategy and application of inclusion criteria yielded 73 eligible studies; six studies for Analysis 1 and 40 studies as subset analyses for Analysis 2. There were another 15 studies for secondary analysis for Analysis 2 where only RR was reported and authors did not provide adequate information to the calculate number of events and patients. The remaining 12 studies were single studies evaluating separate medicine‐gene variant pairs (see Figure 1).

Analysis 1 – PGx (guideline established) versus usual therapy

Of the six studies included in Analysis 1, three were randomized studies ² , ¹⁵ , ¹⁶ and three were non‐randomized studies with either concurrent ¹⁷ or historical controls. ¹⁸ , ¹⁹ Characteristics of the 5 eligible clinical trials for which data could be aggregated are summarized in Table 1. The sixth study was evaluated within the secondary analysis (Table S5). ² Three studies evaluated PGx‐guided dose reduction of fluoropyrimidines to reduce toxicity, two studies evaluated PGx‐guided dose escalation of irinotecan to increase efficacy and reported on toxicities, and one randomized cross‐over cluster trial evaluated optimization of medicine or dose selection for multiple medicines, both anti‐cancer therapies and supportive care medicines, within an oncology cluster to reduce toxicity. There were no other studies that specifically evaluated PGx interventions compared to usual therapy for supportive care medicines such as systemic analgesic opioids or antiemetics.

TABLE 1.

Characteristics of studies included in meta‐analysis for Analysis 1: PGx versus usual therapy.

First author's last name, Year	Study design	Total N recruited	Cancer diagnosis	Summary of PGx intervention and goal	Age range	Geographical regions (countries)
Deenen, 2016 a NCT00838370	Prospective cohort + historical comparator	66	Gastrointestinal, gynecological, and other cancers	1 CPIC/DPWG‐endorsed DPYD variant guided dose reduction or avoidance of fluoropyrimidines to reduce toxicity	44–76	Europe (Netherlands only)
Henricks, 2018 NCT02324452	Prospective cohort + historical comparator	424	Breast, gastrointestinal, and other cancers	4 CPIC/DPWG‐endorsed DPYD variant guided dose reduction or avoidance of fluoropyrimidines to reduce toxicity	21–91	Europe (Netherlands only)
Boisdron‐Celle, 2017 NCT01547923	Prospective cohort + comparator	1116	Locally advanced and metastatic colorectal cancer	4 CPIC/DPWG‐endorsed DPYD variant guided dose reduction or avoidance of fluoropyrimidines to reduce toxicity	24–88	Europe (France only)
Tsai, 2020 NCT02256800	Randomized controlled, parallel arm trial	213	Metastatic colorectal cancer	1 CPIC/DPWG‐endorsed UGT1A1 variant guided dose escalation of irinotecan to improve treatment efficacy	25–83	East Asia (Taiwan only)
Paez, 2019 NCT01639326	Randomized controlled, parallel arm trial	82	Metastatic colorectal cancer	1 CPIC/DPWG‐endorsed UGT1A1 variant guided dose escalation of irinotecan to improve treatment efficacy	40–77	Europe (Spain only)

^a Lunenburg 2018 and Henricks 2019 excluded given the same trial data used and outcomes collected retrospectively.

Analysis 1: Risk of bias for included studies

In general, the quality of included studies was moderate to high, as summarized in Table S2. Authors recognized bias associated with open‐label design of all trials and use of historical controls within some trials with endpoints mostly assessed and graded by unblinded investigators. However, review authors considered it unlikely that primary endpoints of SAEs would be either under‐ or over‐reported in any arm. Consequently, lack of blinding was deemed to be associated with low RoB. RoB scores were allocated for each domain and aggregated as per methods. The overall evidence was qualified as moderate using GRADE (Table 3).

TABLE 3.

Grade Evidence Profile: (Analysis 1, Outcome 1) severe adverse events and (Analysis 1, Outcome 2) treatment efficacy for PGx compared to Usual Therapy, and (Analysis 2, Outcomes 3–11) severe adverse events for emergent pharmacogenomic variant allele carriers compared to wild‐type (WT) carriers in patients receiving treatment for cancer.

Quality assessment							Number of events/patients (%)		Effect		Certainty of the evidence (GRADE)	Importance
№ of studies	Study design	Risk of bias	Inconsistency	Indirectness	Imprecision	Other considerations	PGx or Variant	Usual Therapy or WT	Relative (95% CI)	Absolute
Outcome 1: CPGx reduces severe adverse events compared to usual therapy
5	Clinical trials	Moderate a	Not serious b	Not serious c	Not serious d	Consistent association (with secondary analysis study) e	156/968 (16.1%)	314/924 (34.0%)	RR 0.72 (0.57 to 0.91)	179 fewer events per 1000 patients	⨁⨁⨁◯ Moderate	Critical
Outcome 2: CPGx improves irinotecan efficacy compared to usual therapy
2	Clinical trials	Low f	Not serious g	Serious h	Not serious i	None	104/147 (70.7%)	61/145 (42.1%)	OR 3.33 (2.05–5.41)	286 more objective responses per 1000 patients	⨁⨁◯◯ Low	Limited importance
Outcome 3: TYMS 5′UTR 28base pair repeat (rs45445694) homozygous and heterozygous carriers are predisposed to increased risk of fluoropyrimidine‐associated severe adverse events.
5	Observational studies within clinical trials	Moderate j	Not serious k	Not serious c	Not serious l	Understated effect m	326/711 (45.9%)	189/865 (21.8%)	RR 1.37 (1.15–1.63)	241 more events per 1000 patients	⨁⨁⨁◯ Moderate	Important
Outcome 4: ABCB1 3435C>T (rs1045642) homozygous carriers are predisposed to increased risk of irinotecan‐associated severe adverse events
3	Observational studies within clinical trials	Low n	Not serious o	Not serious c	Not serious p	Strong association q	33/66 (50.0%)	12/67 (17.9%)	RR 2.70 (1.56, 4.67)	321 more per 1000 patients	⨁⨁⨁⨁ High	Important
Outcome 5: UGT1A1*93 (rs10929302) homozygous and heterozygous carriers are predisposed to increased risk of irinotecan‐associated severe adverse events
3	Observational studies within clinical trials	Moderate r	Not serious s	Not serious c	Not serious p	None	29/49 (59.2%)	137/498 (27.5%)	RR 2.08 (1.31, 3.31)	317 more events per 1000 patients	⨁⨁⨁◯ Moderate	Important
Outcome 6: UGT1A1*28 (rs3064744) homozygous carriers are predisposed to an increased risk of severe adverse events from treatment with sacituzumab govitecan
2	Observational studies within clinical trials	Moderate r	Not serious	Not serious	Somewhat t	Moderate association u	85/97 (87.6%)	185/347 (53.3%)	RR 1.65 (1.46, 1.87)	343 more events per 1000patients	⨁⨁⨁◯ Moderate	Important
Outcome 7: HLA‐DRB1*07:01 homozygous and heterozygous carriers are predisposed to increased risk of lapatinib‐associated severe adverse events
2	Observational studies within clinical trials	Moderate r	Not serious	Not serious	Not serious	Strong association v	49/433 (11.3%)	16/1481 (1.1%)	RR 10.36 (5.94, 18.06)	102 more events per 1000patients	⨁⨁⨁◯ Moderate	Important
Outcome 8: UGT1A1*28 homozygous carriers are predisposed to increased risk of nilotinib‐associated severe adverse events
2	Observational studies within clinical trials	Low	Not serious	Not serious	Somewhat w	Strong association x	7/18 (38.9%)	4/49 (8.2%)	RR 4.50 (1.59, 12.74)	307 more events per 1000 patients	⨁⨁⨁◯ Moderate	Important
Outcome 9: ABCB1 c1236T>C (rs1045642) homozygous and heterozygous carriers are predisposed to an increased risk of palbociclib‐associated severe adverse events
1	Observational studies within clinical trials	Low r	Not serious	Not serious c	Not serious	None	58/152 (38.2%)	45/195 (23.1%)	RR 1.65 (1.19, 2.29)	151 more events per 1000 patients	⨁⨁⨁◯ Moderate	Important
Outcome 10: ABCB1 1199G>A rs2229109 homozygous and heterozygous carriers are predisposed to an increased risk of severe adverse events from R‐CHOP/AVCBP like chemotherapy regimens
1	Observational studies within clinical trials	Low r	Not serious s	Not serious c	Somewhat y	Strong association z	13/57 (22.8%)	45/676 (6.7%)	RR 3.53 (2.03, 6.15)	161 more events per 1000patients	⨁⨁⨁◯ Moderate	Important
Outcome 11: OPRM1 118 A>G homozygous and heterozygous carriers are predisposed to greater pain following the commencement of opioid analgesics
4	Observational studies	Serious aa	Serious ab	Not serious	Not serious	None	N/A	N/A	Mean Difference 1.61 (0.77, 2.45)	N/A	⨁⨁◯◯ Low	Important

Abbreviations: CI, confidence interval; OR, odds ratio; RR, risk ratio.

^a RoB: three studies scored moderate and three studies scored low‐risk in the RoB Assessment.

^b Inconsistency: I ² = 34%, 95% Cl of all five individual studies are overlapping; heterogeneity of studies explained with differing aims (treatment efficacy vs. toxicity) and sensitivity analysis conducted.

^c Indirectness: The evidence directly answers the health care question asked.

^d Imprecision: The 95% Cl does not cross one, and criteria for optimal information size is met, that is, total number of patients included is more than the number of patients generated by a conventional sample size calculation for a single adequately powered trial.

^e Other considerations include publication bias, large effect, plausible confounding, and dose‐response gradient. Publication bias has no impact. Effect is consistent across participants with unfavorable genotype.

^f RoB: Two studies recorded low risk in the RoB assessment.

^g Inconsistency: p‐value <0.0001 and I ² = 0%, 95% Cl of both studies are overlapping.

^h Indirectness: Objective response is a surrogate marker of treatment efficacy; overall survival results are conflicting. High‐dose irinotecan is not routinely used in clinical practice.

ⁱ Imprecision: The 95% Cl does not cross one, and criteria for optimal information size is met.

^j RoB: Three studies were low risk; two studies were moderate risk in the RoB assessment.

^k Inconsistency: p‐value <0.01 and I ² = 26%, 95% Cl of all five individual studies are overlapping.

^l Imprecision: the 95% Cl does not cross one and criteria for optimal information size is met.

^m Other considerations include plausible confounding as two studies that showed lower effect size included heterozygous variant carriers in the control arm.

ⁿ RoB: two studies recorded as low‐risk, one study recorded as moderate risk in the RoB assessment.

^o Inconsistency: p‐value <0.001; I ² = 0%. 95% Cl of all three individual studies are overlapping. Although sample size is less than 200, tau square is zero indicating low variability.

^p Imprecision: The 95% Cl does not cross one, and criteria for optimal information size is met.

^q Other considerations include large effect assessed as RR = 2.7 and strength of effects associated with homozygosity. This results in certainty of evidence being upgraded by one level. Despite small sample size (less than 200 patients), funnel plot analysis suggests no publication bias.

^r RoB: all studies recorded as moderate‐risk in the RoB assessment.

^s Inconsistency: p‐value <0.0001; I ² = 0%; 95% Cl of individual studies do not cross 1 and both are overlapping.

^t Imprecision: The 95% Cl does not cross one. Although the total number of patients included in the systematic review is more than the number of patients generated by a conventional sample size calculation for a single adequately powered trial, the variant carrier arm in the systematic review has less patients than would be calculated for single trial (97 vs. 123). This results in certainty of evidence being downgraded by one level.

^u Other considerations. RR = 1.65 (RR between 0.5 and 2.0) showing the effect was moderate.

^v Other considerations include large effect assessed as RR = 10.36. This results in certainty of evidence being upgraded by one level.

^w Imprecision: The 95% Cl does not cross one. Although total number of patients included in the systematic review is more than the number of patients generated by a conventional sample size calculation for a single adequately powered trial, the variant carrier arm in the systematic review has less patients than would be calculated for a single trial (18 vs. 29). This results in certainty of evidence being downgraded by one level.

^x Other considerations include large effect assessed as RR = 4.5. This results in certainty of evidence being upgraded by one level.

^y Imprecision: The 95% Cl does not cross one. Although the total number of patients included in the systematic review is more than the number of patients generated by a conventional sample size calculation for a single adequately powered trial, the variant carrier arm in the systematic review has less patients than would be calculated for a single trial (57 vs. 75). This results in certainty of evidence being downgraded by one level.

^z Other considerations include large effect assessed as RR = 3.4. This results in certainty of evidence being upgraded by one level.

^aa All studies were moderate to high risk in the RoB assessment as these were mostly non‐clinical trial datasets.

^ab Inconsistency: I ² = 91%, 95% Cl of all individual studies do not overlap.

Analysis 1: Effect of intervention on severe adverse events

The meta‐analysis (5 trials, 1892 participants) found lower absolute incidence of SAEs in patients treated with PGx‐guided strategies compared to usual therapy, 16.1% versus 34.0% (RR = 0.72, 95%CI 0.57–0.91, Z = 2.76, I ² = 34%, p = 0.006), Figure 2. Studies were considered homogenous with I ² < 40%. In secondary analysis, the study by Swen et al. showed a strong effect size for the Italy cohort which comprised of oncology patients; OR = 0.24 (95%CI 0.06–1.01) but this study was underpowered and did not achieve statistical significance. ²

FIGURE 2.

Forest plot for Analysis 1: Severe adverse events during the study period for Pharmacogenomics (PGx) guided strategy versus usual therapy.

Three studies reported deaths, with lower incidences of treatment‐related death among DPYD variant allele carriers in the PGx arm compared to Usual Therapy arm (0% vs. 2.5% ¹⁷ ; 0% vs. 10% ¹⁸ ; 0% vs. 8% ¹⁹ ).

Analysis 1; secondary outcome: Effect of intervention on cancer treatment efficacy

Only two of the six studies reporting on safety outcomes also evaluated efficacy outcomes (total 292 participants). The objective disease response rate in the PGx arm versus Usual Therapy arm was 71% versus 42% (RR = 3.33, 95%CI 2.05–5.41, I ² = 0%) indicating a strong association for higher treatment response to irinotecan with UGT1A1‐guided dose escalation strategies. One study showed an overall survival benefit (median 30.0 vs. 22.0 months; p = 0.02) while the other did not demonstrate a statistically significant benefit (26 vs. 17.6 months, p = 0.74).

Analysis 2 (no PGx guideline established): Severe adverse events or symptom burden

Overall, 67 studies reporting SAEs assessed 21 pharmacogenes and 19 variants that are not yet associated with CPIC/DPWG recommendations for medicine optimization. Data from 40 studies were pooled for primary Analysis 2 to evaluate 21 medicine‐gene variant pairs. Twenty‐three studies (6781 patients) were conducted in countries with large European‐ancestry populations; 4 studies (6485 patients) included mixed populations across countries with representation from European, Asian, and some Latin American and African ancestries; 10 studies (819 patients) included patients with East Asian ancestry only, and three studies (398 patients) included patients with Latin American ancestry (182 patients) or Middle Eastern ancestry (216 patients) only. The most studied disease streams were gastrointestinal cancers (19 studies, 4533 patients); breast cancers (8 studies, 8107 patients); mixed solid tumors (7 studies, 1237 patients); and hematological cancers (3 studies, 905 patients). Characteristics of the included clinical trials for which data could be aggregated are summarized in Table 2, and severe adverse events for these studies are summarized in Table S4. The most frequently studied genes were MTHFR (n = 7) for fluoropyrimidines and methotrexate, UGT1A1 (n = 8) for irinotecan (variants other than UGT1A1*28 and UGT1A1*6), nilotinib, sacituzumab govitecan, and etoposide, ABCB1 (n = 9) for irinotecan, palbociclib, and anthracycline‐vinca alkaloid‐cyclophosphamide based therapies, TYMS (n = 5) for fluoropyrimidines, UGT1A6/UGT1A7/UGT1A9 (n = 5) for irinotecan, and HLA‐DRB1 (n = 2) for lapatinib. Ten studies reported on symptom burden changes for pain or vomiting assessed six pharmacogenes: ABCB1, ABCG2, COMT, OPRM1, CYP3A4, and SLC6A4.

TABLE 2.

Characteristics of studies included in meta‐analysis for Analysis 2: SNP carriers versus wild‐type carriers (2 or more studies for individual medicine: pharmacogenomic variant pairs).

First author last name, Year; Trial no. of original clinical study (if known)	Study design	Total N	Medicine Class: Gene pairs of interest	SNPs evaluated (rsID)	Tissue type genotyped	Cancer diagnosis	Treatment regimens	Age (range or mean +/− SD)	Geographical Regions (countries)	Hardy‐Weinberg equilibrium reported and in equilibrium?
Boige, 2010; NCT00126256	Cohort study from Phase III RCT	346	Fluoropyrimidines: TYMS, MTHFR, DPYD Irinotecan: UGT1A1	TYMS 3′UTR 9 bp‐indel (rs11280056) TYMS 5′UTR 28 bp‐repeat (rs45445694) MTHFR c.1298A>C (rs1801131) MTHFR c.677C>T (rs1801133) UGT1A1*93; c.3156G>A (rs10929302)	Peripheral blood	Metastatic colorectal cancer	LV5FU2, FOLFOX6, FOLFIRI	34–83	UK/Europe (France)	Yes
Carlini, 2005	Cohort study from Phase II single‐arm trial	66	Irinotecan: UGT1A9	UGT1A9*22; 118(T)9 > 10 (rs3832043)	Peripheral blood	Metastatic colorectal cancer	Capecitabine/Irinotecan	40–81	North America (US)	Yes
Cote, 2007	Cohort study from Phase III trial	93	Irinotecan: UGT1A1 Irinotecan: ABCB1	UGT1A1*93; c.3156G>A (rs10929302) ABCB1/rs1045642 (3435C>T)	Resected non‐tumor tissue	Stage III colorectal cancer	LV5FU2‐irinotecan	N/A	UK/Europe (France)	Yes
Castro‐rojas, 2017	Single‐arm clinical trial	63	Fluoropyrimidines: TYMS	TYMS 5′UTR 28 bp‐repeat (rs45445694) TYMS 3′UTR 9 bp‐indel (rs11280056); not reported	Peripheral blood	Stage III‐IV colorectal cancer	FOLFOX, XELOX	N/A	Latin America (Mexico)	Yes
Glimelius, 2011	Cohort study from RCT	140	Fluoropyrimidines: MTHFR Irinotecan: ABCB1	MTHFR c.677C>T rs1801133 ABCB1/rs1045642 (3435C>T)	Resected non‐tumor tissue	Stage IV colorectal cancer	FLIRI, LV5FU2‐irinotecan	42–76	UK/Europe (Sweden/Europe)	Yes
Hazama, 2013 UMIN000002388 UMIN000002476	Cohort study from two Phase II RCTs	75	Irinotecan: UGT1A1 Irinotecan: UGT1A9	UGT1A1*60 UGT1A1*93 UGT1A9*1b UGT1A7*3	Peripheral blood	Metatstatic or refractory colorectal cancer	Modified FOLFIRI	N/A	East Asia	Yes
Han, 2009 a	Cohort study from two Phase II RCTs	107	Irinotecan: ABCB1 Irinotecan: UGT1A9 Irinotecan: ABCC2 Irinotecan: SLCO1B1 Irinotecan: CYP3A5	ABCB1 3435C>T (rs1045642) UGT1A9*22; 118(T)9 > 10 (rs3832043) ABCC2 − 24C>T (rs717620), ABCC2 – 1249G>A (rs2273697), and 3972C>T (rs3740066), SLCO1B1 − 11187G>A, 388A > G (rs2306283), and 521T > C (rs4149056), CYP3A5 22893G>A (*3, rs776746)	Peripheral blood	Stage IIIb‐IV non‐small lung cancer	Irinotecan/cisplatin	29–76	East Asia	Yes
Jennings, 2013	Single‐arm clinical trial	254	Fluoropyrimidines: MTHFR Fluoropyrimidines: TYMS	MTHFR c.1298A>C (rs1801131) MTHFR c.677C>T (rs1801133) TYMS 5′UTR 28 bp‐repeat (rs45445694)	Peripheral blood	Colorectal cancer	5‐FU or capecitabine monotherapy or FOLFIRI, CAPIRI, FOLFOX, CAPOX	23–88	UK/Europe	Yes
Kim, 2013 NCT00506571	Phase II single‐arm clinical trial	43	Irinotecan: UGT1A6 Irinotecan: UGT1A7	UGT1A6*2a UGT1A7*3	Peripheral blood	Previously untreated metastatic colorectal cancer	TIROX	30–67	East Asia (Korea)	Not reported
Levi, 2017 NCT00852228	Single‐arm clinical trial	52	Irinotecan: ABCB1	ABCB1 3435C>T (rs1045642)	Peripheral blood	Colorectal cancer with unresectable liver metastases	Hepatic artery infusion of irinotecan, 5‐fluorouracil and oxaliplatin and IV cetuximab	33–76	UK/Europe (multiple)	Yes
Martinez‐Balibrea, 2010	Cohort study of Phase III RCT	149	Irinotecan: UGT1A9 Irinotecan: UGT1A7	UGT1A9*22; 118(T)9 > 10 (rs3832043) UGT1A7*3	Peripheral blood	Metastatic colorectal cancer	FUIRI; FOLFIRI	N/A	UK/Europe	Yes
McLeod, 2010 NCT00003594	Cohort study of Phase III RCT	520	Irinotecan: UGT1A1 Irinotecan: ABCB1	UGT1A1*93; c.3156G>A (rs10929302) ABCB1; 3435C>T (rs1045642)	Peripheral blood	Metastatic colorectal cancer	IFL (fluorouracil ± irinotecan), FOLFOX, and IROX	26–85	North America (US)	Yes
Park, 2011	Phase II single‐arm clinical trial	44	Irinotecan: UGT1A6 Irinotecan: UGT1A7	UGT1A6*2 UGT1A7*3	Peripheral blood	Metastatic gastric cancer	TIROX	27–66	East Asia	Yes
Ruzzo, 2017 NCT00646607	Cohort study of Phase III RCT	534	Fluoropyrimidines: DPYD	DPYD*6; p.V732I; c.2194G>A (rs1801160) DPYD; c.496 A > G (rs2297595)	Peripheral blood	Surgically resected stage III and high‐risk stage II colorectal cancer	FOLFOX‐4 or XELOX adjuvant chemotherapy	57–70 (IQR)	UK/Europe (Italy)	Yes
Schwab, 2008	Single‐arm clinical trial	683	Fluoropyrimidines: MTHFR Fluoropyrimidines: TYMS	MTHFR c.1298A>C (rs1801131) MTHFR c.677C>T (rs1801133) TYMS 5′UTR 28 bp‐repeat (rs45445694)	Peripheral blood	Gastrointestinal cancers, cancer of unknown primary and breast cancer	Varied 5FU regimens	N/A	UK/Europe	Yes (MTHFR) Moderate deviation for TYMS
Seo, 2009	Single‐arm clinical trial	94	Fluoropyrimidines: TYMS	TYMS 5′UTR 28 bp‐repeat (rs45445694)	Peripheral blood	Metastatic or relapsed gastric cancer	Modified FOLFOX, Modified FOLFIRI	N/A	East Asia (Korea)	Not reported
Smyth, 2017	Cohort study of Phase III RCT	124	Fluoropyrimidines: TYMS	TYMS 5′UTR 28 bp‐repeat (rs45445694)	Resected non‐tumor tissue	Resectable gastroesophageal cancer	Perioperative chemotherapy with ECF	29–85	UK/Europe (UK)	Yes
Thomas, 2011	Single‐arm clinical trial	131	Fluoropyrimidines: MTHFR	MTHFR c.1298A>C (rs1801131) MTHFR c.677C>T (rs1801133)	Peripheral blood	Advanced or metastatic rectal cancer	5‐FU ± irinotecan + radiotherapy	N/A	North America (US)	Yes
Boige, 2016 NCT00265811	Cohort study of Phase III RCT	1545	Fluoropyrimidines: DPYD	DPYD*6; c.2194G>A (rs1801160)	Peripheral blood	Stage III Colon Cancer	FOLFOX4 +/− cetuximab	19–75	UK/Europe (France)	Not reported
Bardia, 2021 b NCT01631552	Cohort study of Phase I/II RCT	403	Sacituzumab Govitecan: UGT1A6	UGT1A1*28 (rs3064744)	Peripheral blood	Multiple metastatic cancers including breast, gastrointestinal, lung	Sacituzumab govitecan 8 mg/kg or 10 mg/kg D1, D8 q21d	31–90	North America (US)	Not reported
Rugo, 2022 NCT02574455	Cohort study of Phase III RCT	243	Sacituzumab Govitecan: UGT1A6	UGT1A1*28 (rs3064744)	Peripheral blood	Refractory/relapse metastatic triple negative breast cancer	Sacituzumab govitecan 10 mg/kg D1, D8 q21d	27–82	UK/Europe, North America	Not reported
Schaid, 2014 NCT00374322	Cohort study of Phase III RCT	1104	Lapatinib: HLA‐DRB1	HLA‐DRB1*07:01	Peripheral blood	Early‐stage breast cancer	Lapatinib 1500 mg orally daily	25–87	UK/Europe, Americas, Africa, Australasia	Not reported
Spraggs, 2018 NCT00490139	Cohort study of Phase III RCT	4618	Lapatinib: HLA‐DRB1	HLA‐DRB1*07:01	Peripheral blood	Early‐stage breast cancer	Lapatinib monotherapy, and in combination or sequence with trastuzumab, in addition to concurrent taxane	N/A	Asia, Australasia, Americas, Africa, UK/Europe.	Not reported
Abumiya, 2014	Single‐arm clinical trial	34	Nilotinib: UGT1A1	UGT1A1*28 (rs3064744) UGT1A1*6 (rs4148323)	Peripheral blood	Chronic myeloid leukemia	Nilotinib	31–83	East Asia (Japan)	Not reported
Singer, 2007 NCT00109707	Cohort study of Phase I/II parallel arm NRCT	111	Nilotinib: UGT1A1	UGT1A1*28 (rs3064744)	Peripheral blood	Chronic myeloid leukemia	Nilotinib	55.1 ± 15.7	Asia, Australasia, North America, UK/Europe	Not reported
Toffoli, 2003	Cohort study of single‐arm clinical trial	43	Methotrexate: MTHFR	MTHFR 677TT (rs1801133)	Peripheral blood	Ovarian cancer	Oral MTX 1.25 mg (one‐half tablet of 2.5 mg) every 12 hr for 21 days.	24–77	UK/Europe	Not reported
Choi, 2017	Single‐arm clinical trial	111	Methotrexate: MTHFR	MTHFR 677TT (rs1801133)	Peripheral blood	Primary central nervous system lymphoma	MTX 3500 mg/m2 for 2 hours on day 1, q2w.	17–86	East Asia	Yes
Goey, 2016 NCT00926640	Cohort study of single‐arm clinical trial	25	Belinostat: UGT1A1	UGT1A1*28 (rs3064744) UGT1A1*6 (rs4148323) UGT1A1*60 (rs4124874)	Peripheral blood	Advanced cancer	Belinostat + cisplatin 80 mg/m2 D1 + etoposide 100 mg/m2 D1, D3 and D4	40–78	North America (US)	Yes
Iwata, 2021 NCT01740427, NCT01942135	Cohort study of Phase III RCTs	652	Palbociclib: ABCB1	ABCB1 1236T > C (rs1128503)	Peripheral blood	Advanced breast cancer	Palbociclib D1‐21 q 28 days	N/A	Asia, Australasia, North America, UK/Europe	Yes
Metharom, 2011 [35]	Cohort study of Phase II RCT	33	Gemcitabine: CDA	CDA c.79 A>C (rs2072671)	Peripheral blood	Pancreatic cancer, NCLC	Gemcitabine weekly for 4 weeks	N/A	Australasia	Not reported
Joerger, 2012	Cohort study of single‐arm clinical trial	146	Gemcitabine:CDA	CDA c.79 A>C (rs2072671)	Peripheral blood	NSCLC	Gemcitabine D1, D8 followed by cisplatin or carboplatin D1	37–79	UK/Europe	Yes
Jordheim, 2015 NCT00140660, NCT00140595, NCT00144807, NCT00169143, NCT00144755, and NCT00135499	Cohort study of RCTs	760	Rituximab‐doxorubicin‐vinca alkaloid‐cyclophosphamide based chemotherapy: ABCB1	ABCB1 1199G>A (rs2229109)	Peripheral blood	DLBCL, other NHL	R‐CHOP/R‐ACVBP	18–93	UK/Europe (Belgium France, Switzerland)	Yes
Reyes‐Gibby, 2007 c	Cohort study	207	Morphine: OPRM1 Morphine: COMT	OPRM1 118A > G (rs1799971) COMT Val158Met (rs4680)	Peripheral blood	Multiple cancers: Urology, Breast, Lung, other	Post‐surgery	29–89	UK/Europe (Norway)	Yes
Rakvag, 2008 d	Cohort study	197	Morphine: COMT	COMT Val158Met (rs4680)	Peripheral blood	Multiple cancers: Urology, Lung, Breast, Hematology	Post‐surgery	64 +/−13	UK/Europe (Norway)	Yes
Liu, 2012	Cohort study	96	Tramadol: OPRM1	OPRM1 118A > G (rs1799971)	Peripheral blood	Gastrointestinal cancers; oxaliplatin‐associated neuropathy	Oxaliplatin related neuropathy	N/A	East Asia (China)	Not reported
Pu, 2019 ChiCTR‐PR‐16007775	Cohort study of RCT	59	Oxycodone: OPRM1 Sufentanil: CYP3A4	OPRM1 118A > G (rs1799971) Sufentanil: CYP3A4*1G	Peripheral blood	Gastrointestinal cancers	Post‐surgery	N/A	East Asia (China)	Yes
Babaoglu, 2005	Cohort study	216	Granisetron: ABCB1	ABCB1 3435C>T (rs1045642)	Peripheral blood	Multiple cancers: breast cancer, lymphoma, and lung cancer	Moderately or highly emetogenic chemotherapy	46 +/− 10.7	Middle East (Turkey)	Yes
Tsuji, 2017 e UMIN000009335	Cohort study of Phase III RCT	156	Granisetron: ABCB1 Palonosetron: ABCG2	ABCB1 3435C>T (rs1045642) ABCB1 2677GT > A (rs2032582) ABCG2 421C > A (rs2231142)	Peripheral blood	Advanced cancer	Cisplatin ≥50 mg/m2	31–76	East Asia (Japan)	Yes
Lavanderos, 2019	Cohort study	119	5HT3antagonist: ABCB1	ABCB1 3435C>T (rs1045642)	Peripheral blood	Germinal (seminoma or non‐seminoma) testicular cancer	Bleomycin, cisplatin, etoposide	16–56	Latin America (Chile)	Not assessed

Note: LVF5FU2 = leucovorin 200 mg/m² and 5‐fluorouracil 400 mg/m ² intravenously on day 1, followed by a 46‐h protracted infusion of 5‐fluorouracil 2400 mg/m².

TIROX = S‐1/irinotecan/oxaliplatin: S‐1 40 mg/m² bd. on days 1–14; irinotecan 150 mg/m² plus oxaliplatin 85 mg/m² on Day 1 every 3 weeks.

IROX = irinotecan 200 mg/m² and oxaliplatin 85 mg/m² on Day 1 every 3 week.

FOLFOX = oxaliplatin 85 mg/m² and leucovorin 1, 200 mg/m² and 5‐fluorouracil bolus 400 mg/m2 on day 1, followed by 22‐h infusion of 5‐fluorouracil 600 mg/m².

XELOX = oxaliplatin 135 mg/m² on day 1 and capecitabine 1000 mg/m² twice daily for 14 days.

FOLFIRI = irinotecan 180 mg/m² and leucovorin 400 mg/m², followed by bolus 400–500 mg/m² 5‐fluorouracil, followed by 46‐h infusion of 5‐fluorouracil 2400–3000 mg/m².

FLIRI = irinotecan 180 mg/m² Day 1, bolus 5‐Fluorouracil 500 mg/m² and Leucovorin 60 mg/m² on day 1 and 2.

R‐CHOP = rituximab 37,550 mg/m², doxorubicin 50 mg/m², vincristine 1.4 mg/m² (cap 2 mg), cyclophosphamide 750 mg/m² D1 and prednisolone 100 mg/m² D1‐5.

R‐ACVBP = rituximab 375 mg/m², doxorubicin 75 mg/m² and cyclophosphamide 1200 mg/m² D1, vindesine 2 mg/m² D1&5, bleomycin 10 mg D1&5, prednisone 60 mg/m² D1‐5.

^a Han, 2006 excluded given same trial.

^b Ocean, 2017 excluded given same trial.

^c Klepstad, 2004 excluded given duplicate cohort.

^d Rakvag, 2005 excluded given duplicate cohort.

^e Yokoi, 2018 excluded given same trial.

Analysis 2 (no PGx guideline established): Risk of bias for included studies

The included studies were mostly low to moderate RoB, as summarized in Table S3. Authors recognized bias associated with comparator arms including variant allele carriers (e.g., heterozygous) or where the comparative study did not report Hardy–Weinberg equilibrium. RoB scores were allocated for each domain and aggregated as per methods. The Overall evidence was qualified using GRADE (Table 3).

Fluoropyrimidines: TYMS 5′UTR 28base pair repeat (rs45445694) homozygous and heterozygous

Seven studies ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ , ²⁶ reported on SAEs across 1747 participants, and another study was excluded as authors reported tissue samples were frequently too degraded for TYMS genotyping. ²⁷ Pooled analysis showed a higher incidence of fluoropyrimidine‐associated SAEs in patients homozygous or heterozygous for variant rs45445694 compared to wild‐type, 46.4% versus 24.6% (RR = 1.31, 95%CI 1.12–1.54, I ² = 26%), see Figure 3a. Two studies ²⁴ , ²⁵ were rated as high RoB as the control arm included heterozygotes of variant rs45445694, potentially underestimating the effect size. Pooled data for studies reporting homozygous carriers only was associated with less heterogeneity and higher RR (24.9% vs. 14.3%; RR = 1.8, 95%CI 1.33, 2.44, I ² = 0%). In secondary analysis, Ruzzo et al. ²⁸ reported that despite a low number of SAEs, the risk of vomiting was higher (OR = 8.83; p = 0.049) in carriers of rs45445694, see Table S6.

FIGURE 3.

Forest plots for Analysis 2: Severe adverse events for variant carriers vs. wild‐type carriers. (a) TYMS 5′UTR (rs45445694) homozygous and heterozygous carriers versus wild‐type in patients treated with fluoropyrimidine‐based chemotherapy, (b) DPYD*6 (rs1801160) homozygous and heterozygous carriers versus wild‐type in patients treated with fluoropyrimidine‐based chemotherapy (c) ABCB1 3435C>T (rs1045642) homozygous carriers versus wild‐type in patients treated with irinotecan‐based chemotherapy, (d) UGT1A1*93 (rs10929302) carriers versus wild‐type in patients treated with irinotecan‐based chemotherapy, (e) UGT1A1*28 (rs3064744) carriers versusversus wild‐type in all patients treated with sacituzumab govitecan, (f) HLA‐DRB1*07:01 carriers versus wild‐type in all patients treated with lapatinib, (g) UGT1A1*28 carriers versus wild‐type in all patients treated with nilotinib, (h) Pain scores (mean ± standard deviation) in OPRM1 118A>G carriers versus wild‐type in patients initiated on systemic opioid medication.

Note: (a) Seo, 2009 excluded from the RR calculation due to the high risk of bias (heterozygous carriers were included in the “wild‐type” arm). With the inclusion of Seo 2009, RR = 1.37 (1.12, 1.67), p = 0.002. (c) Hh = Homozygous and heterozygous carriers (Han 2009; McLeod 2010) excluded from RR calculation. (d) Event rate per patient year used for McLeod 2010; RR not significantly different in sensitivity analysis. (e) Event rate per patient year used for Rugo 2022; RR not significantly different in sensitivity analysis.

Fluoropyrimidines: DPYD*6 (rs1801160) homozygous and heterozygous

Two studies ²⁹ , ³⁰ reported on SAEs across 2042 participants. Pooled analysis showed that both homozygous and heterozygous carriers of DPYD*6 allele were at higher risk of fluoropyrimidine‐associated SAEs, 60.0% vs. 44.6% (RR = 1.42; 95%CI = 1.10–1.84, I ²  = 73%), see Figure 3b.

Irinotecan: ABCB1 3435C>T (rs1045642) homozygous

Five studies ²³ , ²⁷ , ³¹ , ³² , ³³ reported on SAEs across 343 participants. Overall, only 3 studies reported outcomes for homozygotes versus wild‐type, ²⁷ , ³¹ , ³³ wherein pooled analysis showed a significantly higher incidence of irinotecan‐associated SAEs in patients homozygous for variant rs1045642 compared to wild‐type, 50% versus 18% (2.70, 95%CI 1.56–4.67), (I ² = 0%), see Figure 3c, with homozygous carriers accounting for 23% of participants. In secondary analysis, Lara et al reported a similar summary effect size. ³⁴ Statistical significance was lost when heterozygous carriers were included in the pooled analysis.

Irinotecan: UGT1A1 *93 (rs10929302) homozygous

Three studies ²⁰ , ²³ , ³¹ (394 participants) in aggregate demonstrated higher incidence of irinotecan‐associated SAEs in homozygous carriers of the UGT1A1*93 (−3156G>A; rs10929302) variant compared to wild‐type, 59.1% versus 27.5% (RR = 2.08, 95%CI 1.31–3.31, I ² = 64%), see Figure 3d. Approximately 8% of participants were homozygous carriers. Mcleod ²³ reported a greater number of events than patients. Accordingly, data for this study was converted to an event rate per patient year for pooled analysis. In secondary analysis, Innocenti et al. also found that ‐3156G>A was a significant predictor of severe neutropenia and distinguished different phenotypes within TA indel genotypes. ³⁵

Sacituzumab govitecan: UGT1A1 *28 (rs3064744) homozygous

Two studies ³⁶ , ³⁷ (370 participants) demonstrated higher incidence of SAEs in homozygous carriers of UGT1A1*28 (rs3064744) variant compared to wild‐type, 87.6% versus 53.3% (RR = 1.55 95%CI 1.46–1.87, I ² = 0%), Figure 3e. As Rugo ³⁷ reported a greater number of events than patients, data was converted to an event rate per patient years for pooled analysis.

Lapatinib: HLA‐DRB1 *07:01 homozygous and heterozygous

Two studies ³⁸ , ³⁹ (1914 participants) demonstrated higher incidence of lapatinib‐associated SAEs, in particular, hepatotoxicity among carriers of HLA‐DRB1*07:01 variant compared to wild‐type, 11.3% versus 1.1% (RR = 10.36, 95%CI 5.94–18.06, I ² = 0%), Figure 3f.

Nilotinib: UGT1A1 *28 (rs3064744) homozygous

Two small studies ⁴⁰ , ⁴¹ (67 participants) demonstrated higher incidence of nilotinib‐associated SAEs, among homozygous carriers of UGT1A1*28 variant compared to wild‐type, 39% versus 8.2% (RR = 4.5, 95%CI 1.59–12.74, I ² = 0%), Figure 3g.

Opioids: OPRM1 118 A>G

Four studies ⁴² , ⁴³ , ⁴⁴ , ⁴⁵ (423 participants) demonstrated higher pain scores among carriers of the OPRM1 118A>G variant compared to wild‐type, with a mean difference of 1.61 (95%CI 0.77–2.45, p = 0.0002, I ² = 91%), Figure 3h. Heterogeneity was considerable as indications for opioid use varied from procedure‐related pain to cancer pain. In secondary analysis, a validation cohort of approximately 620 patients receiving opioids for moderate to severe cancer pain, unrelated to a procedure, showed no association between this variant and pain intensity. ⁴⁶

Palbociclib: ABCB1 rs1128503 homozygous and heterozygous

One study ⁴⁷ (347 participants) reported on the interactions between palbociclib and variants of the ABCB1 gene, c1236C>T (rs1128503) and c3435C>T (rs1045642). A higher incidence of palbociclib‐associated SAEs occurred among homozygous and heterozygous carriers of the c1236C>T variant compared to wild‐type, 38% versus 23% (RR = 1.65 95%CI 1.19–2.29, p = 0.003) and 32% versus 23% (RR = 1.37 95%CI 1.03–1.84, p = 0.03) respectively but not c3435C>T (rs1045642) variant. Of the participants studied, 23% were homozygous carriers of c1236C>T and 70% were homozygous or heterozygous carriers. The interaction of c1236C>T variant on palbociclib was significant in non‐Asian patients (32% vs. 19%) but not in Asians (50% vs. 67%).

R‐CHOP/AVCBP: ABCB1 rs2229109 homozygous and heterozygous

In one study ⁴⁸ of 760 participants, a higher incidence of SAEs for non‐Hodgkin lymphoma patients receiving chemotherapy with regimen R‐CHOP (rituximab, cyclophosphamide, doxorubicin, vincristine, prednisolone) or R‐AVCBP (rituximab, doxorubicin, cyclophosphamide, vindesine, bleomycin, prednisone) occurred among homozygous and heterozygous carriers of the ABCB1 1199G>A (rs2229109) variant compared to wild‐type, 22.8% versus 6.5% (RR = 3.53; 95%CI 2.03–6.15, p < 0.0001). Approximately, 8% of participants carried this variant allele. In secondary analysis, Rossi et al ⁴⁹ reported a higher risk of severe cardiac toxicity in heterozygous carriers compared to wild‐type; OR = 1.89 (95%CI 1.15–3.12, p = 0.01).

Other medicine: variant pairs

Meta‐analysis from two studies ⁵⁰ , ⁵¹ totaling 166 participants showed higher incidence of gemcitabine‐associated SAEs among homozygous and heterozygous carriers of the CDA c79 A>C (rs2072671) variant compared to wild‐type, 36.4% versus 16.7% (RR = 2.25, 95%CI 1.13–4.47, I ² = 16%). However, secondary analysis demonstrated no correlation between this CDA genotype and occurrence of toxicities, ⁵² and a conflicting effect with this variant associated with a lower risk of severe toxicities. ⁵³

Pairs that did not show a significant effect on incidence of SAEs included fluoropyrimidines: MTHFR rs1801133 (5 studies, ²⁰ , ²² , ²⁴ , ²⁷ , ⁵⁴ 754 participants), RR = 0.89, 95%CI 0.6–1.3, I ² = 0%; fluoropyrimidines: MTHFR rs1801131 (4 studies, ²⁰ , ²² , ²⁴ , ⁵⁴ 625 participants), RR = 1.05, 95%CI 0.65–1.70, I ² = 0% (and secondary analysis ⁵⁵ ); irinotecan: UGT1A7*3 (3 studies, ⁵⁶ , ⁵⁷ , ⁵⁸ 190 participants), RR = 1.15, 95%CI 0.77–1.74, I ² = 0%; irinotecan: UGT1A6*2 (2 studies, ⁵⁶ , ⁵⁸ 86 participants), RR = 1.8, 95%CI 0.26–12.48, I ² = 53%; irinotecan: UGT1A9*1b (4 studies, ³² , ⁵⁷ , ⁵⁹ , ⁶⁰ 394 participants), RR = 0.91, 95%CI 0.57–1.43, I ² = 61%, and pazopanib: HLA‐B*57:01 (1 study, ⁶¹ 1002 participants, RR = 1.14, 95%CI 0.55–2.36).

Two studies ⁶² , ⁶³ that evaluated MTHFR rs1801133 in patients treated with methotrexate (362 participants), each utilizing very different dosing regimens, were associated with a higher incidence of SAEs compared to wild‐type, 39% versus 24.3%, but this was not a statistically significant result in pooled analysis (RR = 1.8 95%CI 0.81–3.99; I ² = 24%). For 5‐HT3 antagonists used as antiemetics, there was no statistically significant effect on loss of emesis control associated with ABCB1 rs1045642 (40% variant carriers vs. 29.5% wild‐type; RR = 1.37, 95% CI 0.46–4.07, I ² = 65%) in pooled analysis of 3 studies ⁶⁴ , ⁶⁵ , ⁶⁶ (407 participants) or ABCG2 rs2231142 ⁶⁶ , ⁶⁷ (20.0% variant carriers vs. 24.3% wild‐type; RR = 1.17, 95% CI 0.08–16.93). Secondary analysis included assessments of association of taxane‐related toxicities with ARHGEF10, ⁶⁸ FGD4 ⁶⁸ , ⁶⁹ ; and everolimus with CYP3A4/5 ⁷⁰ Data could not be pooled to evaluate medicine: variant pairs that were associated with single and inadequately powered studies including imatinib: ABCB1 1236T>C rs1128503, ⁷¹ high dose etoposide: UGT1A1*28, ⁷² sorafenib: CYP3A5, ⁷³ sorafenib: ABCB1 3435C>T rs1045642, ⁷⁴ belinostat: UGT1A1*28, ⁷⁵ sunitinib: ABCG2 421 C>A rs2231142 ⁷⁶ (including secondary analysis ⁷⁷ ), taxanes: ATP‐binding cassette family of genes, ⁷⁸ , ⁷⁹ (including secondary analysis ⁸⁰ , ⁸¹ ); taxanes: CYP3A4/5, ⁸² (including secondary analysis ⁸³ , ⁸⁴ , ⁸⁵ , ⁸⁶ ); pralatrexate: ATIC, ⁸⁷ morphine: SLC6A4, ⁸⁸ vincristine: CEP72. ⁸⁹ These results are shown in Tables S4 and S5.

DISCUSSION

This meta‐analysis demonstrates that for the medicine‐gene pairs that have been subject to study in adult patients with cancer, adoption of evidence‐based pharmacogenomic‐guided prescribing significantly reduces the incidence of clinically relevant, adverse medicine‐related events including chemotherapy‐related deaths. Our secondary analysis suggests the potential for greater impact when comprehensive and multigene testing is employed as part of the PGx strategy, in support of large‐scale PGx implementation approaches across multiple jurisdictions, including but not limited to cancer care. These studies have also demonstrated feasibility and acceptance in clinical practice. ² , ⁹⁰ , ⁹¹ , ⁹² , ⁹³ Here, we recommend, for priority clinical validation within a cancer care context, eight pharmacogenomic variants that are not currently associated with a recommendation for medicines optimization from CPIC or DPWG guidelines but have been found to be predictive of a higher risk of SAEs compared to wild‐type allele carriers. These will be studied in a prospective clinical trial ACTRN12624000079549.

The clinical applicability for these emergent variants is important, as assessed by GRADE, given adverse events were prospectively collected and limited to SAEs only, analysis was generally restricted to clinical trial datasets where pharmacovigilance requirements ensure robust follow‐up of SAEs, and the effect sizes were generally moderate or large. The same regimens were used across intervention/at‐risk arms and comparator arms with standard dose adjustments applied depending on toxicities; and differences in dose–response gradients were considered as part of the generalisability assessments for GRADE. Only studies that prospectively replicated or validated a PGx association were included, and not PGx discovery‐only cohorts or studies that may utilize proximal surrogate outcomes such as pharmacokinetic changes or pharmacodynamic biochemical responses. It is well‐reported that false positive findings are typically associated with discovery studies that have design limitations or test many potential associations without proper statistical correction for multiple comparisons. ⁹⁴ Finally, studies that genotyped tumor tissue were excluded given that somatic variants may be acquired during carcinogenesis and prior reports have shown conflicting data when tumor tissue was utilized for pharmacogenomic analysis. ⁹⁵ Although most included studies utilized peripheral blood for genotyping, a minority of studies utilized fresh frozen normal tissue, and where authors ²⁷ reported low quality of sampled tissue for variant analysis, these data were excluded. Notably, in limiting our included studies to robust clinical datasets, less studied and more novel variants with important associations may not have been identified through this approach.

The present review demonstrated that DPYD‐guided dosing of fluoropyrimidines reduced SAEs and treatment‐related deaths as shown in previously published reviews of diverse study design. ⁹⁶ Although the included studies did not assess the impact on treatment efficacy, a newer study has shown no detrimental impact on survival for many variants ⁹⁷ and further research is underway to optimize the accuracy of fluoropyrimidine dosing. ⁹⁸ , ⁹⁹ We highlight a small effect size on the incidence of severe toxicity among homozygous and heterozygous carriers of DPYD*6, which supports findings from a recent meta‐analysis of any grade fluoropyrimidine‐associated toxicity. ¹⁰⁰ The DPWG and CPIC have regarded this variant as having a negligible effect on clinically relevant enzymatic degradation of fluoropyrimidines based on studies conducted largely in European countries. However, the minor allele frequency of this variant is highest in some South Asian populations (9%) where increased toxicity has been suggested in small studies, ¹⁰¹ , ¹⁰² and in African populations where it has been associated with reduced enzymatic activity conditional on the presence of a linked variant p.Y186C. ¹⁰³ This review also supports findings from a previously published meta‐analysis associating TYMS 5′UTR with an increased risk of adverse events. ¹⁰⁴ Combined with DPYD, concurrent TYMS genotyping may enhance the predictability of patients at elevated risk for adverse events. ¹⁰⁵ This is particularly important where target concentration intervention modeling cannot be utilized to guide dosing as in the case of the oral prodrug capecitabine where intracellular concentrations of 5‐fluoroucail cannot be routinely measured nor have been correlated with systemic concentrations of capecitabine. ⁹⁹

This meta‐analysis also showed that in addition to patients who are carriers of UGT1A1*28 (rs3064744) and UGT1A1*6 (rs4148323) variants, patients who have the UGT1A1*93 (rs10929302), and ABCB1 3435 C>T (rs1045642) variant alleles are at significantly increased risk of severe toxicity from irinotecan‐based therapies. SN‐38, a DNA topoisomerase I inhibitor, is the active metabolite of irinotecan. It is inactivated by intracellular glucuronidation via the uridine diphosphate glucuronosyltransferase (UGT) family of enzymes including UGT1A1. Both UGT1A1*28 and UGT1A1*6 variants, recognized by the CPIC and DPWG, diminish glucuronidation and increase exposure to SN‐38 with patients who are homozygous for either or both variants requiring dose reduction of irinotecan to avoid toxicities. UGT1A1*93 has been reported to be in linkage disequilibrium with *28 in populations of European‐ancestry ¹⁰⁶ and therefore may not have an independent effect on irinotecan toxicity, however, these findings have not been confirmed across multiple ethnicities. The ABCB1 encoded transporter, ABCB1 or P‐glycoprotein actively expels irinotecan, SN‐38, and its glucuronide out of hepatocytes into the bile canaliculus lumen and intestinal lumen. Homozygous carriers of ABCB1 3435C>T (rs1045642) may be predisposed to increased risk of toxicity through reduced efflux ability of ABCB1 and reduced clearance of SN‐38. ¹⁰⁷ The minor allele frequency is highest in African (78%) and Asian (62%) ancestries and lowest in European (48%) and South Asian (39%) ancestries. ¹⁰¹ While other variants such as ABCB1 1236C>T (rs1128503) have previously been correlated with increased exposure to irinotecan and its active metabolite SN‐38, ¹⁰⁸ our analysis did not uncover an association of severe toxicities within a prospective clinical trial dataset. The influence of ABCB1 variants on irinotecan‐induced severe toxicity may be better elucidated through haplotype analysis. ¹⁰⁹ These medicine‐variant interactions may also be relevant for the third‐generation antibody‐drug conjugate, sacituzumab govitecan which is comprised of monoclonal antibody hRS7 conjugated to SN38 via a hydrolyzable linker. Although available data suggest homozygous UGT1A1*28 carriers are predisposed to increased toxicity from sacituzumab govitecan (65.0% vs. 46.6%; RR = 3.57, 95%CI 1.83–6.95) similar to irinotecan, further analysis is warranted to determine if other variants of relevance to irinotecan, namely UGT1A1*6, UGT1A1*93 and ABCB1 3435 C>T are equally important for toxicities related to sacituzumab govitecan.

The ABCB1 gene was shown in this review to be an important predictor of toxicity across a range of anti‐cancer regimens including the more novel Cyclin‐dependent kinases 4 and 6 inhibitors. Despite evidence of improved overall survival from CDK 4/6 inhibitors in hormone‐receptor positive, human epidermal growth factor receptor 2 (HER2)–negative advanced breast cancer, wide inter‐individual variability in therapeutic benefit and toxicity have been reported. The ABCB1 1236C>T variant was associated with higher rates of toxicity from palbociclib in non‐Asians (32% vs. 19%) but not in Asians. While an exposure‐safety relationship for palbociclib has previously been reported with higher palbociclib exposure predisposing to a significantly higher risk of neutropenia and thrombocytopenia, there is no evidence that an exposure‐response relationship at the standard fixed dose exists. ¹¹⁰ , ¹¹¹ Inter‐ethnic differences in tolerability of small molecule anti‐cancer agents have previously been correlated to ethnic differences in the expression and activity of metabolic enzymes and transporters including for the P‐glycoprotein. ¹¹² Ethnicity was not well‐reported across included studies and if reported, it was race that was generally self‐identified by participants. Most studies were conducted in European‐ancestry population‐rich geographical regions, followed by East Asian countries. The landscape for pharmacogenomic mapping across geographical regions has improved and enabled a more accurate determination of genomic variation across other ethnic populations including for Indigenous peoples. ¹¹³ , ¹¹⁴ , ¹¹⁵ , ¹¹⁶ Investigation of ancestry‐specific differences in linkage disequilibrium will elucidate lesser‐known variants that are of greater impact in poorly studied populations and highlight the differential impact on drug response by ancestry. ¹¹⁷

There are a few limitations of this meta‐analysis. First, there is heterogeneity in the types of studies evaluated under Analysis 1 (PGx vs. Usual Therapy) where the included studies aimed to either enhance cancer treatment efficacy or reduce cancer treatment toxicity, and different strategies, both single variant testing and broad gene panel testing were evaluated. Regardless of the treatment goal of the individual studies, overall, PGx is demonstrated to improve safety, and where maintaining dose intensity of chemotherapy is important for disease control, avoidance of toxicities will minimize chemotherapy dose delays or interruption and may have an overall impact on treatment outcomes. Furthermore, a subset analysis was conducted to clarify the impact on clinically relevant safety outcomes. Second, criteria were not concordant for Analysis 2 (no PGx guideline recommendation) where CTCAE criteria were used to define adverse events but symptom burden was defined by prespecified measures of brief pain inventory and emesis control rates. Accordingly, Analysis 2 comprised of subset analyses for individual medicine‐variant pairs where there was a limited number of good quality studies that could be aggregated for assessment of individual medicine: variant effects. Where genotype and outcomes data were not available for aggregation, those studies provided a narrative synthesis to either corroborate or contest the effect estimates from pooled analyses.

CONCLUSION

In conclusion, this comprehensive analysis of all available English language studies confirms the clinical validity of CPIC/DPWG‐endorsed pharmacogenomics‐guided approaches for medicines optimization to reduce unwarranted and preventable SAEs in patients with cancer. We advocate for future studies evaluating the pharmacogenomics of anti‐cancer therapies to include efficacy endpoints and not just safety endpoints. Additionally, eight emerging pharmacogenomic variants are identified as priorities for future research, prior to consideration for implementation into clinical practice.

AUTHOR CONTRIBUTIONS

All authors wrote the manuscript. S.Li. and C.M.J.K. designed the research; S.Li., S.La., M.S., C.M.J.K., and J.N. performed the research. All authors analyzed the data.

FUNDING INFORMATION

No funding was received for this work.

CONFLICT OF INTEREST STATEMENT

SLo receives research funding from institutions from Novartis, Bristol Myers Squibb, MSD, Puma Biotechnology, Eli Lilly, Nektar Therapeutics, Astra Zeneca, and Seattle Genetics. She has acted as a consultant (not compensated) to Seattle Genetics, Novartis, Bristol Myers Squibb, MSD, AstraZeneca, Eli Lilly, Pfizer, Gilead Therapeutics, and Roche‐Genentech. She has acted as a consultant (paid to institution) to Aduro Biotech, Novartis, GlaxoSmithKline, Roche‐Genentech, Astra Zeneca, Silverback Therapeutics, G1 Therapeutics, PUMA Biotechnologies, Pfizer, Gilead Therapeutics, Seattle Genetics, Daiichi Sankyo, MSD, Amunix, Tallac Therapeutics, Eli Lilly and Bristol Myers Squibb. JFS receives an honorarium from AbbVie, Astra Zeneca, Beigene, BMS, Genor Bio, Gilead, Janssen, Roche, and TG Therapeutics for participation on advisory boards, and research funding from Abbvie, BMS, Janssen, and Roche. None of these Sponsors were involved in the review. All other authors declared no competing interests for this work.

CONSENT FOR PUBLICATION

All authors have reviewed the final manuscript and provided consent for publication.

PROTOCOL REGISTRATION

The study methods were specified in advance and documented in a study protocol. The study protocol was registered with PROSPERO, registration number CRD42023421277. Prospero protocol is publicly accessible. Minor amendments to the protocol were made after the commencement of the review justified by preliminary searches, screening, and data review, to refine the inclusion criteria.

REPORTING GUIDELINES

Systematic reviews and meta‐analyses (PRISMA) checklist provided in Supplementary Tables.

Supporting information

Table S1.

Lingaratnam S, Shah M, Nicolazzo J, et al. A systematic review and meta‐analysis of the impacts of germline pharmacogenomics on severe toxicity and symptom burden in adult patients with cancer. Clin Transl Sci. 2024;17:e13781. doi: 10.1111/cts.13781

DATA AVAILABILITY STATEMENT

All data generated or analyzed during this study are included in this published article [and its supplementary information files].