Dataset Definitions, Outcome Definition, and Case/Control Counts
All patients were classified as opioid-naive or opioid-exposed and had a binary outcome for postsurgical persistent opioid use (POU) (Fig. 1). Our full sample comprised 3,198 (8.1%) POU cases and 36,321 controls; our opioid-naive subset comprised 794 (2.4%) cases and 32,656 controls (Table 1). In both cohorts, the median age of participants was 55 years and slightly fewer than half of participants (46.5%) were male; approximately half of the cases (full: 53.1%; naive: 49.9%) and one-third of the controls (full: 36.7%; naive: 35.1%) had Michigan Prescription Drug Monitoring Program data (Table 1).
Table 1
Cohort description. Breakdown of age, sex, and prescription source for both the full MGI dataset and its subset, the opioid-naive dataset. Percentages denote proportions within each column. PDMP = Prescription Drug Monitoring Program (in Michigan).
| Full Sample | Opioid-Naive Sample |
| Overall | Cases | Controls | Overall | Cases | Controls |
N | 39,519 | 3,198 | 36,321 | 33,450 | 794 | 32,656 |
Median age (sd) | 55 (16.3) | 56 (13.7) | 55 (16.5) | 55 (16.5) | 56 (15.1) | 55 (16.6) |
No. male (%) | 18,384 (46.5) | 1,411 (44.1) | 16,973 (46.7) | 15,544 (46.5) | 353 (44.4) | 15,191 (46.5) |
No. PDMP (%) | 15,032 (38.0) | 1,699 (53.1) | 13,333 (36.7) | 11,858 (35.4) | 396 (49.9) | 11,462 (35.1) |
Survey of Previous Studies
Our survey of previous studies encompassed 72 peer-reviewed publications between 1997 and 2022 that reported genetic associations with OUD or related outcomes (Fig. 2; Supplementary Table 1). Among 98 study populations in these publications, the majority had either European (n = 30) or Asian (n = 29) ancestry, 19 were trans-ancestry, and 11 had African ancestry; three studies did not specify the predominant ancestry of participants (Fig. 2). Of the analyses producing significant associations in the 72 studies, 54 (75.0%) were candidate gene/SNP, 13 (18.1%) were GWAS, one (1.9%) was gene-based,65 and four (7.4%) had a combination of candidate, GWAS, and/or gene-based analyses (Fig. 1). Fifty of the studies (69.4%) had OUD outcomes, including heroin dependence, “opioid addiction,” prior morphine use, and DSM-IV criteria for opioid dependence; of these OUD outcomes, 36 (72.0%) exclusively studied heroin addiction (Fig. 1; Supplementary Table 1). Most other studies were part of clinical pain research, drug treatments, and pharmacogenetic research (Fig. 2). For analyses with binary outcomes, the median case count was 459 and the median control count was 340; for quantitative outcomes, the median sample size was 112 (Fig. 2; Supplementary Fig. 1).
The genes that appeared in the most candidate studies were OPRD1 (n = 12), OPRM1 (n = 10), and DRD2/ANKK1 (n = 6) (Fig. 3). These genes also had the largest number of unique SNPs appearing in candidate studies: DRD2/ANKK1 (n = 16), OPRD1 (n = 15), and OPRM1 (n = 10) (Fig. 3). When considering all studies, the greatest number of analyses producing significant associations among genes occurred in OPRM1 (25 analyses across 15 studies) followed by OPRD1 and DRD2/ANKK1; among SNPs the greatest number occurred with rs1799971-OPRM1 (15 analyses across nine studies) followed by rs1800497-DRD2/ANKK1 and rs4680-COMT (Supplementary Fig. 2).
Genetic Association Candidate Study
When applied to 72 previous studies with OUD or related outcomes, our candidate SNP inclusion criteria (Methods) returned 77 unique SNPs across 23 genes to test for association with POU in MGI. Eleven out of 15 previously reported high-value variants in OPRM1, the gene encoding the µ-opioid receptor, had nominally significant associations (p < 0.05) in MGI (Table 2; Supplementary Fig. 2). Six of these previously appeared with significant associations, as defined by each analysis, in candidate studies only (most significant in MGI: rs3778150-C, ORfull=1.14, pfull=2.9×10− 4) (Fig. 3; Table 2).19,45,46 The remaining five had at least one significant association (p ≤ 1.5×10− 8) in a previous GWAS (most significant in MGI: rs79704991-T, ORfull=1.17, pfull=8.7×10− 5) (Fig. 3; Table 2).2,10,11 The four unreplicated OPRM1 variants (rs2075572, rs495491, rs510769, and rs6848893) appeared in candidate studies only.13,19,22 The MGI POU phenotype was also nominally associated with three other SNPs previously found in candidate studies: rs10886472-A-GRK5 (ORnaive=0.89, pnaive=0.028), rs4633-T-COMT (ORfull=1.07, pfull=0.017), and rs4680-A-COMT (ORfull=1.07, pfull=0.016) (Fig. 3; Table 2; Supplementary Material).55,66
Table 2
MGI nominally significant (p < 0.05) associations at previously reported variants associated with OUD or related outcomes. INFO lists the rsID, effect allele, and nearest gene of each discovery variant. DISCOVERY DATASET lists information pertaining to previously reported associations. MGI indicates the persistent opioid use p-values, effect sizes, and relevant datasets. Bolded values indicate pMGI<0.0022.
SNP INFO | DISCOVERY DATASET | | MGI PERSISTENT OPIOID USE |
rsID (effect allele) | Nearest Gene | Outcome | Effect (type) | P-value | Sample Size | No. Cases | No. Controls | Ancestry | Analysis Type | Reference | Effect (OR) | P-value | MAF | Dataset |
rs3778151 (T) | OPRM1 | Opioid use disorder | -6.487 (beta) | 8.8 × 10–11 | 447,580 | 37,826 | 409,754 | European; African; Hispanic | GWAS meta-analysis | Kember 2022 | 0.89 0.81 | 5.6 × 10 − 4 1.9 × 10 − 3 | 0.17 0.17 | Full Naive |
rs3778149 (C) | OPRM1 | Opioid use disorder | -6.086 (beta) | 1.2 × 10 − 9 | 321,312 | 24,866 | 296,446 | European | GWAS meta-analysis | Kember 2022 | 0.88 0.82 | 3.5 × 10 − 4 2.9 × 10 − 3 | 0.17 0.17 | Full Naive |
rs1799971 (A) | OPRM1 | Opioid use disorder | 7.315 (beta) | 2.6 × 10–13 | 321,312 | 24,866 | 296,446 | European | GWAS meta-analysis | Kember 2022 | 1.11 | 9.5 × 10 − 3 | 0.13 | Full |
rs79704991 (T) | OPRM1 | Opioid use disorder | 5.71 (beta) | 1.5 × 10 − 8 | 554,186 | 15,251 | 538,935 | European | GWAS meta-analysis | Deak 2022 | 1.17 1.31 | 8.7 × 10 − 5 5.1 × 10 − 4 | 0.13 0.13 | Full Naive |
rs9478500 (C) | OPRM1 | Opioid addiction | 0.136 (beta) | 2.6 × 10 − 9 | 304,831 | 7,281 | 297,550 | European | GWAS meta-analysis | Gaddis 2022 | 1.14 1.23 | 3.3 × 10 − 4 2.4 × 10 − 3 | 0.17 0.17 | Full Naive |
rs3823010 (A) | OPRM1 | Heroin addiction | 1.23 (OR) | 2.7 × 10 − 8 | 16,729 | 4,287 | 12,442 | European; African | Candidate meta-analysis | Hancock 2015 | 1.13 1.23 | 4.1 × 10 − 4 2.9 × 10 − 3 | 0.17 0.17 | Full Naive |
rs511435 (T) | OPRM1 | Heroin addiction | 1.16 (OR) | 3.0 × 10 − 6 | 16,729 | 4,287 | 12,442 | European; African | Candidate meta-analysis | Hancock 2015 | 1.07 1.17 | 2.7 × 10 − 2 1.1 × 10 − 2 | 0.22 0.22 | Full Naive |
rs524731 (A) | OPRM1 | Heroin addiction | 1.16 (OR) | 1.2 × 10 − 6 | 16,729 | 4,287 | 12,442 | European; African | Candidate meta-analysis | Hancock 2015 | 1.07 1.17 | 3.0 × 10 − 2 1.3 × 10 − 2 | 0.22 0.22 | Full Naive |
rs3778150 (C) | OPRM1 | Heroin addiction | 1.28 (OR) | 1.2 × 10 − 4 | 10,703 | 1,950 | 8,753 | European; African | Candidate meta-analysis | Hancock 2015 | 1.14 1.24 | 2.9 × 10 − 4 1.7 × 10 − 3 | 0.17 0.17 | Full Naive |
rs562859 (C) | OPRM1 | Heroin addiction | 1.12 (OR) | 4.3 × 10 − 3 | 10,703 | 1,950 | 8,753 | European; African | Candidate meta-analysis | Hancock 2015 | 1.11 | 4.8 × 10 − 2 | 0.34 | Naive |
rs9479757 (A) | OPRM1 | Daily heroin dose | 3.45 (Ridit) | < 0.01 | 392 | 232 | 160 | Han Chinese | Candidate | Shi 2002 | 0.87 0.79 | 2.3 × 10 − 3 6.6 × 10 − 3 | 0.09 0.09 | Full Naive |
SNP INFO | DISCOVERY DATASET | | MGI PERSISTENT OPIOID USE |
rsID (effect allele) | Nearest Gene | Outcome | Effect (type) | P-value | Sample Size | No. Cases | No. Controls | Ancestry | Analysis Type | Reference | Effect (OR) | P-value | MAF | Dataset |
rs10886472 (A) | GRK5 | Methadone dosage (mg) | 15.86 (beta) | 0.0003 | 344 | --- | --- | Taiwanese | Candidate | Wang 2018 | 0.89 | 0.028 | 0.47 | Naive |
rs4633 (T) | COMT | Pharmacodynamic effects of butorphanol on ulna pressure pain threshold (kg) | 0.37 (beta) | 0.015 | 108 | --- | --- | European; African | Candidate | Ho 2019 | 1.07 | 0.017 | 0.49 | Full |
rs4680 (A) | COMT | Morphine-induced changes in ischemic pain threshold (seconds) | -16.70 (beta) | 0.041 | 108 | --- | --- | European; African | Candidate | Ho 2019 | 1.07 | 0.016 | 0.49 | Full |
The MGI POU phenotype did not have any nominally significant associations with variants in the three other opioid receptor genes: OPRD1 (n = 9), OPRK1 (n = 4), and OPRL1 (n = 2) (Fig. 3). All studies producing associations in these genes were candidate analyses with sample sizes < 3,500 (median: 288) and case counts < 1,500 (median: 202) (Fig. 3; Supplementary Table 1).5–7,14,16,18,20,22,25,40,42–44,52–54 Similarly, POU was not significantly associated with any variants in DRD2/ANKK1 (n = 9), all of which were discovered in candidate studies with relatively small case counts (median = 1,459) and control counts (median = 349).17,40,47,48 In total, there were no meaningful associations at 63 of the 77 high-value SNPs, of which 54 (85.7%) were originally found in candidate analyses only (Fig. 3). As MGI’s power to replicate increased there was an increase in the proportion of variants with nominal significance and consistent direction with results from the literature survey; nonetheless, among 46 SNPs with 91–100% replication power, only 38 (82.6%) had consistent direction of effect and only 16 (34.8%) replicated the literature results (Fig. 3).