Genetic risk signatures of opioid-induced respiratory depression following pediatric tonsillectomy

Abstract

Background: Respiratory depression is a clinically and economically important but preventable complication of opioids. Genetic factors can help identify patients with high risk for respiratory depression. Methods: In this prospective genotype blinded clinical study, we evaluated the effect of a panel of variants in candidate genes on opioid-related respiratory depression in 347 children following tonsillectomy. Results: Using unsupervised hierarchical clustering and a combination of candidate genotypes and clinical variables, we identified several distinct clusters of patients at high risk (36-38%) and low risk (10-17%) of respiratory depression; the relative risk of respiratory depression for high versus low risk clusters was 2.1-3.8 (p = 0.003). Conclusion: Genetic risk predictions (genetic signatures) along with clinical risk factors effectively identify children at higher and lower risks of opioid-induced respiratory depression. Genetic signatures of respiratory depression offer strategies for improved clinical decision support to guide clinicians to balance the risks of opioid adverse effects with analgesia. Original submitted 9 July 2014; Revision submitted 19 September 2014.

Keywords: children; depression; genetic signature; morphine; opioid; pediatric tonsillectomy; pharmacogenetics; respiratory; surgical pain.

Figures

Figure 1. The consort diagram

Illustrates the flow of study participants through this clinical trial. Eligible participants, reasons for exclusions, enrolled and analyzed patients are reported. IRB: Institutional review board.

Figure 2. Correlation of total morphine dose with race, sex, obstructive sleep apnea and respiratory depression

In the discovery cohort, total morphine dose (mg/kg) was correlated with race, sex (female and male), OSA (no or yes) and RD (no or yes). Children with OSA received overall less total morphine. Higher total morphine dose was associated with RD. Standard box plots are used with horizontal bars corresponding to median values, boxes and whiskers representing two middle quartiles and 1.5 of the interquartile ranges, respectively. Outliers are denoted by circles. AA: African–American; C: Caucasian; O: Other race; OSA: Obstructive sleep apnea; RD: Respiratory depression.

Figure 3. Gene signatures of varying respiratory depression risks

Genetic variant signatures identify high respiratory depression risk versus low respiratory depression risk subtypes among children requiring postsurgical opioid interventions in the recovery room. Clustering of children in the discovery cohort was done using a set of candidate SNPs and clinical variables as features; Hamming distance was used to define similarity. The main high and low respiratory depression clusters were subsequently subdivided into five clusters with progressively increasing risk of respiratory depression. Percentage of respiratory depression incidence in each cluster is shown.

Figure 4. Hierarchical clustering of respiratory depression in children based on underlying genetic variants and clinical factors

Overall hierarchical (bi-) clustering of patients (columns) and SNPs/clinical variables (rows) for a subset of patients who needed additional intervention (DCint cohort). In the heat map, green, black and gray are used to indicate different possible values for SNP and other variables. For each cluster, risk is indicated by the color of a box delineating that cluster: high risk by red, intermediate risk by orange and low risk by blue boxes, respectively. The number of respiratory depression cases and the overall number of subjects in each cluster are indicated in each box. SNPs contributing most to the clustering pattern (rs2295632, rs441417, rs3766246, rs324420 and rs1045642) are indicated by a horizontal purple oval. (A) Results for the entire DCint cohort (same as in Figure 3 – note that that the two main clusters are flipped), using SNP genotypes and several clinical variables. (B) Results for a subset of patients of Caucasian descent (DCint-cau), using SNP genotypes only. Please see color figure at www.futuremedicine.com/doi/pdf/10.2217/pgs.14.137

Figure 5. Recursive partitioning respiratory depression decision tree

Recursive partitioning of children in the discovery cohort who received additional opioid interventions in recovery room (DCint) for their risk of respiratory depression (yes = high risk; no = low risk) using a combination of three top SNPs, rs2295632, rs1045642 and rs1042713. FAAH SNP, rs2295632 was the most discriminative for high versus low respiratory depression risk. These three SNPs were also found to contribute significantly to the clustering patterns shown in Figures 3 & 4. In Figure 5, these SNPs are combined into a simple logical rule that classifies children into low- versus high-risk strata ‘rs2295632_A = N’ denotes the lack of A allele, that is, CC genotype, as opposed to rs2295632 = Y, which corresponds to CA and AA genotypes that contain an A allele.

Figure 6. Crossvalidation of respiratory depression risk prediction using k-nearest neighbors

Nearest neighborhood crossvalidation of respiratory depression (RD) risk distribution of the number of RD neighbors in the leave-one-out crossvalidation among seven nearest neighbors for RD (B & D) versus no RD cases (A & C) in the training cohort (A & B) and in the control set (C & D). The corresponding distributions for non-RD cases are shifted to the left, enabling prototype-based classification of high RD risk cases found to be similar (as defined by the Hamming distance in the space of SNP and clinical variables used for Figure 4) to other RD cases.