Predicting individual differences in placebo analgesia: contributions of brain activity during anticipation and pain experience

Abstract

Recent studies have identified brain correlates of placebo analgesia, but none have assessed how accurately patterns of brain activity can predict individual differences in placebo responses. We reanalyzed data from two fMRI studies of placebo analgesia (N = 47), using patterns of fMRI activity during the anticipation and experience of pain to predict new subjects' scores on placebo analgesia and placebo-induced changes in pain processing. We used a cross-validated regression procedure, LASSO-PCR, which provided both unbiased estimates of predictive accuracy and interpretable maps of which regions are most important for prediction. Increased anticipatory activity in a frontoparietal network and decreases in a posterior insular/temporal network predicted placebo analgesia. Patterns of anticipatory activity across the cortex predicted a moderate amount of variance in the placebo response (∼12% overall, ∼40% for study 2 alone), which is substantial considering the multiple likely contributing factors. The most predictive regions were those associated with emotional appraisal, rather than cognitive control or pain processing. During pain, decreases in limbic and paralimbic regions most strongly predicted placebo analgesia. Responses within canonical pain-processing regions explained significant variance in placebo analgesia, but the pattern of effects was inconsistent with widespread decreases in nociceptive processing. Together, the findings suggest that engagement of emotional appraisal circuits drives individual variation in placebo analgesia, rather than early suppression of nociceptive processing. This approach provides a framework that will allow prediction accuracy to increase as new studies provide more precise information for future predictive models.

Figures

Figure 1.

Overview of analyses and methods. A, A schematic showing which relationships were tested in each of analyses 1–5. B, Standard multiple regression, in which each voxel is treated as a separate outcome measure, and multiple regression is used to assess the effects of predictors of interest (e.g., placebo analgesia) and covariates on brain activity in that voxel. C, LASSO-PCR, in which multiple brain voxels and covariates are entered in a single regression model as predictors of the outcome. Observations (subjects) are split into training and test data in a cross-validation scheme. Training data are used to build the regression model and estimate voxel weights (regression slopes), and test data are used to assess prediction accuracy. For the training data, PCA is performed on the contrast maps of ranked placebo − control contrast values. Using LASSO regression, component scores (V) are regressed on placebo analgesia scores, and a set of weights (β, regression slopes) are obtained. The purpose of both PCA and LASSO is to reduce the impact of multicollinearity, thus increasing prediction accuracy and yielding stable (low-variance) weights on voxels that are neuroscientifically interpretable. Finally, the contrast data are extracted for test subjects and the model applied to obtain a predicted outcome (e.g., placebo analgesia) score.

Figure 2.

Anticipatory predictors of placebo analgesia from standard regression. A, Surface figures (middle) illustrate regions where greater placebo related increases (warm colors) and decreases (cool colors) during pain anticipation predicted greater placebo analgesia. Scatter plots (left) illustrate the relationship between average cluster activity during anticipation (y-axis) and reported placebo analgesia (x-axis). Examples are shown for the left DLPFC (positive correlation) and for the right SII region (negative correlation). BOLD activity reported on the y-axis ascends from top to bottom, because activity is displayed relative to placebo condition, where the contrast values were negative when P > C and positive when C > P. The slices (right) show in greater detail that increases predicted analgesia in the VLPFC and OFC, while decreases predicted analgesia in SII and a region encompassing the insula and superior temporal gyrus (STG). B, Regions in which anticipatory activity was correlated with greater prescan expectations of analgesia in study 2. The scatter plot shows the negative correlation in pgACC/rmPFC, averaged across voxels.

Figure 3.

Anticipatory predictors of placebo analgesia, comparison across methods. A, Surface figures showing results for standard regression (analysis 1). B, Grouping of significant voxels in analysis 1 into functional networks (see supplemental material, available at www.jneurosci.org, for details). Colors indicate distinct groups of intercorrelated regions. C, Weights (regression slopes) reliably predictive of placebo analgesia in the LASSO-PCR analysis (FDR q < 0.05, p < 0.016). Predictions for new subjects are obtained by multiplying the entire weight map (not only the colored regions) by the subject's contrast values. The most reliably predictive voxels (colored) aid in interpretation of the functional anatomy of predictive systems. Yellow/red colors indicate positive effects, greater placebo analgesia with greater placebo − control increases. Blue colors indicate negative effects, greater analgesia with reduced placebo − control activity (or placebo-related deactivation).

Figure 4.

Predictive strength of anticipatory activity: unbiased estimates using LASSO-PCR. A, Predictive accuracy obtained from LASSO-PCR within an a priori cortical mask for studies 1 (blue) and 2 (red). The scatter plot shows observed placebo-analgesia scores (y-axis) versus predicted by the cross-validated analyses (x-axis). The accuracy estimate is minimally biased; the cross-validation scheme is known to slightly underestimate predictive accuracy. B, Minimally biased predictions from LASSO-PCR for study 2 alone. Predictive accuracy was substantially stronger for study 2.

Figure 5.

Direct comparison of standard regression (analysis 1) and LASSO-PCR voxel weights (analysis 2). A, Overlap between standard regression weights in analysis 1 and LASSO-PCR results from analysis 2 at the same threshold (p < 0.016). Positive results unique to LASSO-PCR are shown in red, those unique to the standard regression in yellow, and the conjunction in orange. Negative results unique for LASSO-PCR are shown in dark blue, unique for standard regression in light blue, and the overlap in medium blue. Overlapping voxels were found in nearly all major regions of activation, except the pregenual cingulate/ventromedial prefrontal effect unique to LASSO-PCR. B, A scatter plot of voxel weights (blue dots) for LASSO-PCR versus standard regression. The black squares indicate the average standard regression weights within 30 evenly spaced LASSO-PCR weight bins, and the dashed lines show ±1 SD. The colored squares show the joint weights for voxels in regions significant in the LASSO-PCR analysis (at least 50 contiguous voxels), with ±1 SD error bars. The plot shows a strong and linear relationship between regression weights for the two techniques, with results in key regions in line with the overall relationship across voxels. C, Two-dimensional histogram of p values for LASSO-PCR (x-axis) versus standard regression (y-axis). The colored contours indicate the log number of voxels with p values in that region of the joint space, with higher voxel counts in red and lower ones in blue. The dashed shows the unity relationship (equal p values) between maps. The predominance of low p value voxels above the unity line indicates lower p values for the LASSO-PCR regression. Colored squares show p values for the regions plotted in B. Though all LASSO-PCR p values for the key regions shown are necessarily low (they were chosen based on the LASSO-PCR results), some regions showed much higher standard regression p values than others, indicating a masked relationship with placebo analgesia that is uncovered in LASSO-PCR by controlling for other brain regions.

Figure 6.

LASSO-PCR based accuracy for predicting placebo analgesia from anticipatory activity within three a priori sets of regions. A, Regions identified in each meta-analysis. Blue, Consistent activations related to executive working memory in a meta-analysis of 60 studies (Wager and Smith, 2003). Red, Consistent activations related to emotional tasks in a meta-analysis of 163 studies (Kober et al., 2008). Yellow, Results from a mega-analysis of 114 participants who underwent a thermal pain challenge in our laboratory (the pain-processing network localizer; see supplemental Fig. 4, available at www.jneurosci.org as supplemental material). B, Prediction accuracy from LASSO-PCR for each set of regions, compared with the “null” model based only on training-set placebo analgesia scores. Only the emotion-related regions showed an above-chance (*p < 0.05) reduction in prediction error. C, Prediction (x-axis)–outcome (y-axis) scatter plots for each set of regions.

Figure 7.

Anticipatory activity predicting placebo effects in average PPN responses from LASSO-PCR (analysis 4). A, Prediction (x-axis)–outcome (y-axis, average [placebo − control] contrast within PPN) scatter plots for each study. B, Map of predictive voxel weights. Colors are as in Figure 3. C, Overlap and dissociations between regions predictive of placebo analgesia in analysis 2 and average placebo effects in PPN (analysis 4). Positive results unique to placebo analgesia are shown in red, those unique to effects in PPN in yellow, and the conjunction in orange. Negative results unique for placebo analgesia are shown in dark blue, unique for effects in PPN in light blue, and the overlap in medium blue. Effects were largely non-overlapping, except in pre-SMA, left inferior frontal gyrus, and cerebellum, suggesting that placebo analgesia and placebo effects in PPN reflect qualitatively distinct anticipatory processes. D, Scatter plot of LASSO-PCR weights predicting reported analgesia (x-axis) versus those predicting PPN effects (y-axis). Colors and lines are as in Figure 5B. This plot shows an essentially null relationship between voxel weights for the two outcomes. CB, Cerebellum; APFC, anterior prefrontal cortex; see the text for other abbreviations.

Figure 8.

Activity during peak pain (placebo − control) predicting placebo analgesia, from LASSO-PCR (analysis 5). A, Prediction (x-axis) versus outcome (y-axis, placebo analgesia) scatter plots for each study. B, The left panel shows the PPN mask that constituted the set of predictors, and the right panel shows the most predictive voxel weights (p < 0.001). Blue indicates areas in which larger placebo-induced decreases during pain predicted placebo analgesia. Yellow/orange indicates areas in which smaller decreases or relative placebo-induced increases predicted placebo analgesia. C, Overlap and dissociations between regions associated with placebo effects in PPN in the LASSO-PCR analysis versus standard regression. Positive results unique to LASSO-PCR are shown in red, those unique to standard regression in yellow, and the conjunction in orange. Negative results unique for LASSO-PCR are shown in dark blue, unique for standard regression in light blue, and the overlap in medium blue. Effects overlapped in posterior cingulate (relative decreases predicted analgesia) and anterior insula (relative increases predicted analgesia), but effects in dorsal cingulate and posterior insula, among other regions, were unique to the LASSO-PCA solution.