PAC 6 report
Date of PAC: 2021-04-12
Date last code update: 2021-04-05
Date last ran: 2021-04-11
Analysis Overview
Test and replication samples. We have finalized a ‘test’ and ‘replication’ sample from the SPINS dataset. The test sample is comprised of n=135 participants scanned on the CMH GE scanner (89 SSD, 87 men). The replication sample is comprised of n=173 participants scanned on the combined CMP, MRP, and ZHP PRISMA scanners (91 SSD, 103 men).
Participant characteristics. Participant characteristics, by sample, are similar. Both samples’ SSD participants have lower education and WTAR. SSD participants show lower performance on all nonsocial and social cognition tests (with the exception of TASIT 1 in the GE sample, and TASIT 2 sincere in the GE and PRISMA sample). Group differences between white matter tracts are less consistent: the PRISMA sample generally distinguishes SSD and HC, but the GE sample is more mixed.
| Variable | Combined | HC | SSD | p |
|---|---|---|---|---|
| Demographics | ||||
| Sex (F:M) | 48:87 | 20:26 | 28:61 | .233 |
| Age (years) | 27.51 (7.68) | 27.17 (8.34) | 27.69 (7.36) | .715 |
| Handedness (0=L, 1=R) | 0.67 (0.45) | 0.70 (0.41) | 0.65 (0.47) | .538 |
| Education (years) | 14.19 (2.13) | 15.48 (1.82) | 13.52 (1.97) | .000 |
| Parental education | 15.73 (2.96) | 16.26 (2.49) | 15.44 (3.15) | .130 |
| WTAR | 113.04 (11.78) | 116.00 (9.07) | 111.52 (12.73) | .036 |
| Neurocognition (MATRICS MCCB) | ||||
| Processing speed | 45.38 (12.52) | 52.22 (9.87) | 41.84 (12.31) | .000 |
| Attention and vigilance | 41.41 (11.26) | 47.80 (9.29) | 38.10 (10.80) | .000 |
| Working memory | 45.22 (11.97) | 52.17 (10.21) | 41.63 (11.25) | .000 |
| Verbal learning | 44.35 (9.58) | 49.15 (8.78) | 41.87 (9.06) | .000 |
| Visual learning | 45.01 (11.28) | 50.04 (7.70) | 42.42 (11.97) | .000 |
| Reasoning and problem solving | 44.64 (9.61) | 47.48 (8.85) | 43.17 (9.71) | .013 |
| Social cognition | ||||
| ER-40 | -2038.66 (573.77) | -1756.65 (275.21) | -2184.42 (632.07) | .000 |
| RMET | 26.51 (4.48) | 28.33 (3.35) | 25.57 (4.71) | .001 |
| EA | 0.84 (0.23) | 0.93 (0.13) | 0.79 (0.26) | .001 |
| RAD | 56.27 (9.43) | 61.26 (5.85) | 53.69 (9.91) | .000 |
| TASIT 1 | 24.19 (2.78) | 24.72 (2.06) | 23.91 (3.06) | .110 |
| TASIT 2 sincere | 17.18 (2.81) | 17.33 (2.99) | 17.10 (2.73) | .662 |
| TASIT 2 simple sarcasm | 16.28 (4.13) | 18.13 (2.04) | 15.33 (4.60) | .000 |
| TASIT 2 paradoxical sarcasm | 17.22 (3.30) | 18.67 (2.07) | 16.47 (3.56) | .000 |
| TASIT 3 lies | 26.12 (4.20) | 28.22 (2.67) | 25.03 (4.44) | .000 |
| TASIT 3 sarcasm | 25.47 (4.58) | 27.33 (3.82) | 24.52 (4.66) | .001 |
| White matter | ||||
| AF left | 0.60 (0.02) | 0.61 (0.02) | 0.60 (0.03) | .730 |
| AF right | 0.57 (0.02) | 0.57 (0.03) | 0.57 (0.02) | .875 |
| CB left | 0.50 (0.03) | 0.50 (0.02) | 0.50 (0.03) | .467 |
| CB right | 0.49 (0.03) | 0.49 (0.03) | 0.48 (0.03) | .072 |
| ILF left | 0.51 (0.02) | 0.52 (0.02) | 0.51 (0.02) | .114 |
| ILF right | 0.51 (0.02) | 0.52 (0.02) | 0.51 (0.02) | .651 |
| IOFF left | 0.63 (0.03) | 0.63 (0.03) | 0.63 (0.02) | .993 |
| IOFF right | 0.63 (0.02) | 0.63 (0.02) | 0.62 (0.02) | .464 |
| UF left | 0.48 (0.03) | 0.48 (0.03) | 0.49 (0.03) | .536 |
| UF right | 0.46 (0.03) | 0.46 (0.03) | 0.46 (0.03) | .917 |
| TF left | 0.49 (0.01) | 0.50 (0.01) | 0.49 (0.01) | .122 |
| TF right | 0.48 (0.01) | 0.49 (0.01) | 0.48 (0.01) | .049 |
| CC1 | 0.51 (0.03) | 0.51 (0.02) | 0.51 (0.03) | .635 |
| CC2 | 0.57 (0.02) | 0.57 (0.02) | 0.57 (0.02) | .350 |
| CC3 | 0.59 (0.02) | 0.60 (0.02) | 0.59 (0.02) | .043 |
| CC4 | 0.60 (0.02) | 0.60 (0.01) | 0.60 (0.02) | .047 |
| CC5 | 0.60 (0.02) | 0.60 (0.02) | 0.60 (0.02) | .819 |
| CC6 | 0.61 (0.01) | 0.62 (0.01) | 0.61 (0.01) | .026 |
| CC7 | 0.61 (0.02) | 0.62 (0.02) | 0.61 (0.02) | .210 |
| Function outcome, health, clinical | ||||
| BSFS | 151.09 (27.02) | 171.17 (21.29) | 140.71 (23.66) | .000 |
| CIRS-G | 2.54 (2.27) | 1.91 (1.35) | 2.87 (2.57) | .020 |
| BPRS | – | – | 29.96 (6.65) | – |
| SAS | – | – | 22.58 (12.09) | – |
| QLS | – | – | 79.77 (20.03) | – |
| SANS | – | – | 2.57 (2.23) | – |
| Medication | ||||
| CPZE equivalents (mg / day) | – | – | 321.96 (290.86) | – |
| Variable | Combined | HC | SSD | p |
|---|---|---|---|---|
| Demographics | ||||
| Sex (F:M) | 70:103 | 41:41 | 29:62 | .023 |
| Age (years) | 32.65 (10.56) | 33.06 (10.48) | 32.27 (10.67) | .626 |
| Handedness (0=L, 1=R) | 0.67 (0.44) | 0.61 (0.50) | 0.72 (0.38) | .090 |
| Education (years) | 14.99 (2.38) | 16.13 (2.10) | 13.96 (2.14) | .000 |
| Parental education | 15.42 (3.07) | 15.69 (2.76) | 15.17 (3.34) | .272 |
| WTAR | 109.28 (12.28) | 113.48 (11.11) | 105.32 (12.10) | .001 |
| Neurocognition (MATRICS MCCB) | ||||
| Processing speed | 45.86 (14.52) | 54.26 (10.09) | 38.30 (13.75) | .000 |
| Attention and vigilance | 43.88 (14.22) | 47.55 (15.29) | 40.58 (12.36) | .001 |
| Working memory | 44.65 (11.52) | 46.95 (12.08) | 42.57 (10.64) | .012 |
| Verbal learning | 45.14 (11.13) | 50.93 (10.33) | 39.93 (9.11) | .000 |
| Visual learning | 42.36 (12.84) | 48.02 (11.02) | 37.26 (12.26) | .000 |
| Reasoning and problem solving | 46.30 (11.61) | 49.57 (10.04) | 43.35 (12.19) | .000 |
| Social cognition | ||||
| ER-40 | -2108.40 (608.06) | -1848.78 (374.13) | -2342.35 (681.08) | .000 |
| RMET | 25.82 (4.92) | 27.13 (4.07) | 24.63 (5.32) | .001 |
| EA | 0.73 (0.30) | 0.83 (0.26) | 0.63 (0.30) | .000 |
| RAD | 55.83 (8.45) | 60.18 (5.61) | 51.91 (8.68) | .000 |
| TASIT 1 | 23.41 (3.14) | 24.79 (2.25) | 22.16 (3.31) | .000 |
| TASIT 2 sincere | 17.13 (3.07) | 17.49 (2.53) | 16.81 (3.46) | .149 |
| TASIT 2 simple sarcasm | 16.83 (3.99) | 18.62 (1.97) | 15.22 (4.63) | .000 |
| TASIT 2 paradoxical sarcasm | 16.77 (3.58) | 18.29 (2.26) | 15.40 (3.99) | .000 |
| TASIT 3 lies | 25.40 (4.66) | 26.38 (4.14) | 24.53 (4.94) | .009 |
| TASIT 3 sarcasm | 25.36 (5.03) | 27.63 (3.10) | 23.32 (5.54) | .000 |
| White matter | ||||
| AF left | 0.53 (0.04) | 0.53 (0.04) | 0.52 (0.04) | .015 |
| AF right | 0.52 (0.04) | 0.53 (0.03) | 0.51 (0.04) | .002 |
| CB left | 0.44 (0.03) | 0.44 (0.03) | 0.43 (0.03) | .005 |
| CB right | 0.43 (0.03) | 0.43 (0.03) | 0.42 (0.03) | .002 |
| ILF left | 0.47 (0.03) | 0.48 (0.03) | 0.46 (0.03) | .000 |
| ILF right | 0.47 (0.03) | 0.48 (0.03) | 0.46 (0.03) | .000 |
| IOFF left | 0.54 (0.04) | 0.55 (0.04) | 0.53 (0.04) | .001 |
| IOFF right | 0.55 (0.04) | 0.55 (0.03) | 0.54 (0.04) | .001 |
| UF left | 0.43 (0.03) | 0.44 (0.03) | 0.43 (0.03) | .686 |
| UF right | 0.44 (0.03) | 0.44 (0.03) | 0.44 (0.03) | .690 |
| TF left | 0.46 (0.02) | 0.46 (0.02) | 0.45 (0.02) | .009 |
| TF right | 0.46 (0.02) | 0.46 (0.02) | 0.46 (0.02) | .006 |
| CC1 | 0.47 (0.03) | 0.48 (0.02) | 0.46 (0.03) | .004 |
| CC2 | 0.53 (0.03) | 0.54 (0.02) | 0.53 (0.03) | .000 |
| CC3 | 0.55 (0.03) | 0.56 (0.02) | 0.55 (0.03) | .001 |
| CC4 | 0.57 (0.03) | 0.57 (0.02) | 0.56 (0.03) | .003 |
| CC5 | 0.57 (0.03) | 0.58 (0.02) | 0.57 (0.03) | .022 |
| CC6 | 0.57 (0.02) | 0.58 (0.02) | 0.57 (0.02) | .002 |
| CC7 | 0.58 (0.02) | 0.59 (0.02) | 0.58 (0.02) | .002 |
| Function outcome, health, clinical | ||||
| BSFS | 154.06 (29.94) | 175.79 (19.10) | 134.47 (23.80) | .000 |
| CIRS-G | 2.89 (3.09) | 1.55 (1.97) | 4.10 (3.42) | .000 |
| BPRS | – | – | 31.16 (7.88) | – |
| SAS | – | – | 25.96 (12.07) | – |
| QLS | – | – | 69.03 (20.57) | – |
| SANS | – | – | 2.32 (2.63) | – |
| Medication | ||||
| CPZE equivalents (mg / day) | – | – | 523.11 (444.68) | – |
CCA analysis. We conducted separate CCAs on the test and replication samples. We elected to include 35 variables in the CCA analysis. In both samples, the \(X\) set is comprised of FA estimates in 19 white matter tracts, representing association, projection, and commissural fibers (pictured below, bilateral association and projection), and the \(Y\) set is comprised of 16 nonsocial and social cognition variables, from the the MATRICS MCCB and SPINS battery, respectively. Thus, the test set has an ~3.86:1 observation:feature ratio, and the replication set has an ~4.94:1 observation:feature ratio.
CCA results
[1] Canonical correlations and redundancies
Both samples show similar canonical correlation \((R_c)\) values across the derived variates. Specifically, the first variate for GE shows \(R_{cTEST}\) = 0.71 (shared variance between variates, \(R_c^2\) = 0.5), and for PRISMA, \(R_{cREPLICATION}\) = 0.72 (\(R_c^2\) = 0.52). Permutation testing found canonical correlation values to be stable, across all variates, in both samples. These high \(R_c\) values reflect the strong linear relationship between participant-wise \(X\) and \(Y\) scores (derived by multiplying the standardized values of each participant’s original variables by the standardized canonical coefficients, and adding the products). The visual on the right shows participant scores on the first variate.
In addition to canonical correlations, it is recommended to calculate redundancy \((R_d)\). Redundancy estimates the extent to which the variance of one set of variables can be accounted for by the other. Note that redundancies are set-specific, i.e., they are not symmetrical like the \(R_c\) values). We calculate the total canonical redundancy for the variables in the \(Y\) set (i.e., the total percent of \(Y\) set variance accounted for by the \(X\) set variables, through all canonical variates: \(R_{dTEST}\) = 21.99%, and \(R_{dREPLICATION}\) = 22.96%.
GE
PRISMA
[2] Statistical testing
We performed a parametric and nonparametric significance test, reporting the four primary test statistics for multivariate models. The parametric tests are not reliably significant for the GE sample, but all parametric test are significant in the PRISMA sample. The permutation tests are all significant for both samples , with the exception of the Pillai’s trace statistic in the GE sample. In all cases of model significance, only the first variate (CV1) is significant. Moreover, in both samples, the first variate’s redundancy warrants interpretation (\(R_{dTEST}\)=10%, \(R_{dREPLICATION}\)=14%).
GE
| variates | test statistic | statistic | critical | df1 | df2 | parametric p | permuted p |
|---|---|---|---|---|---|---|---|
| 1-16 | Wilks | 0.052 | 1.143 | 304 | 1274.432 | 0.065 | 0.05 |
| 1-16 | Hotelling | 3.655 | 1.180 | 304 | 1570.000 | 0.027 | 0.036 |
| 1-16 | Pillai | 2.471 | 1.105 | 304 | 1840.000 | 0.119 | 0.108 |
| 1-16 | Roy | 0.502 | 7.427 | 16 | 118.000 | 0 | 0.016 |
Below, we visualize the permuted Hotelling-Lawley trace statistic, which is significant in both the parametric and nonparamatric significance tests, in both the GE and PRISMA samples.
PRISMA
| variates | test statistic | statistic | critical | df1 | df2 | parametric p | permuted p |
|---|---|---|---|---|---|---|---|
| 1-16 | Wilks | 0.067 | 1.411 | 304 | 1741.384 | 0 | 0 |
| 1-16 | Hotelling | 3.335 | 1.493 | 304 | 2178.000 | 0 | 0 |
| 1-16 | Pillai | 2.264 | 1.327 | 304 | 2448.000 | 0 | 0 |
| 1-16 | Roy | 0.516 | 10.411 | 16 | 156.000 | 0 | 0 |
Below, we visualize the permuted Hotelling-Lawley trace statistic, which is significant in both the parametric and nonparamatric significance tests, in both the GE and PRISMA samples.
[3] Variate structure
Standardized structure coefficients (\(r_s\)) indicate the variables that were important to forming the canonical variate from that variable’s set. They can be interpreted like regression coefficients, i.e., a one unit increase in the given variable would result in an increase/decrease of the \(rs\) value. By convention, (\(r_s\)) values ≥ ~.45 are interpreted.
Though permutation testing found structure coefficients to be consistent within the GE and PRISMA samples, we see here that the variate structure is somewhat inconsistent between the GE and PRISMA samples (see also the visualization in the comparison table tab). In general, the PRISMA sample has more highly contributing variables: only three variables contribute in the GE sample but not in the PRISMA (left ILF to the \(X\) set, and working memory and ER-40 to the \(Y\) set); eight variables contribute in the PRISMA sample but not the GE (right AF, right TF, and left UF to the \(X\) set, and EA, RAD, TASIT 1, TASIT 2 paradoxical sarcasm, and TASIT 2 simple sarcasm to the \(Y\) set). We interpret this pattern of results across both the GE and PRISMA samples as follows: the significant association between the \(X\) and \(Y\) variates is driven by the CC3 and right UF, several neurocognitive domains, and higher-order social cognition.
GE
| Variable | \(\beta\) | \(r_s\) | \(r_s^2\) (\(h^2\)) |
|---|---|---|---|
| X | |||
| AF_left | -0.284 | 0.152 | 0.023 |
| AF_right | 0.283 | -0.216 | 0.046 |
| CB_left | 0.060 | -0.026 | 0.001 |
| CB_right | -0.285 | -0.318 | 0.101 |
| CC1_commissural | -0.247 | -0.207 | 0.043 |
| CC2_commissural | 0.511 | -0.05 | 0.002 |
| CC3_commissural | -0.405 | -0.501 | 0.251 |
| CC4_commissural | -0.261 | -0.428 | 0.183 |
| CC5_commissural | 0.108 | -0.066 | 0.004 |
| CC6_commissural | -0.055 | -0.045 | 0.002 |
| CC7_commissural | -0.028 | -0.074 | 0.006 |
| ILF_left | 0.313 | 0.588 | 0.345 |
| ILF_right | 0.270 | -0.102 | 0.010 |
| IOFF_left | -0.403 | -0.082 | 0.007 |
| IOFF_right | 0.334 | 0.094 | 0.009 |
| TF_left | 0.619 | 0.032 | 0.001 |
| TF_right | -0.752 | -0.445 | 0.198 |
| UF_left | -0.016 | -0.222 | 0.049 |
| UF_right | -0.165 | -0.592 | 0.350 |
| Y | |||
| attn_vig | -0.147 | -0.631 | 0.398 |
| problem_solving | 0.195 | -0.212 | 0.045 |
| process_speed | -0.101 | -0.561 | 0.315 |
| verbal_learning | 0.129 | -0.47 | 0.221 |
| visual_learning | -0.501 | -0.665 | 0.442 |
| work_memory | -0.378 | -0.657 | 0.432 |
| EA | -0.269 | -0.414 | 0.171 |
| ER_40 | -0.400 | -0.493 | 0.243 |
| RAD | 0.242 | -0.425 | 0.181 |
| RMET | 0.099 | -0.259 | 0.067 |
| TASIT_1 | 0.379 | -0.036 | 0.001 |
| TASIT_2_paradoxicalSarcasm | 0.220 | -0.299 | 0.089 |
| TASIT_2_simpleSarcasm | -0.260 | -0.387 | 0.150 |
| TASIT_2_sincere | 0.144 | 0.11 | 0.012 |
| TASIT_3_lies | -0.015 | -0.227 | 0.051 |
| TASIT_3_sarcasm | -0.326 | -0.46 | 0.211 |
| Note: | |||
|
\(r_s\) values > .45 are emphasized, following convention; \(\beta\) = standardized canonical function coefficient; \(r_s\) = structure coefficient; \(r_s^2\) = squared structure coefficient, here also communality \(h^2\) |
|||
PRISMA
| Variable | \(\beta\) | \(r_s\) | \(r_s^2\) (\(h^2\)) |
|---|---|---|---|
| X | |||
| AF_left | 0.270 | 0.085 | 0.007 |
| AF_right | -0.048 | -0.529 | 0.280 |
| CB_left | -0.227 | -0.279 | 0.078 |
| CB_right | 0.051 | -0.447 | 0.200 |
| CC1_commissural | 0.187 | -0.12 | 0.014 |
| CC2_commissural | 0.215 | -0.39 | 0.152 |
| CC3_commissural | -0.664 | -0.635 | 0.404 |
| CC4_commissural | 0.228 | -0.449 | 0.201 |
| CC5_commissural | -0.079 | -0.237 | 0.056 |
| CC6_commissural | 0.316 | -0.084 | 0.007 |
| CC7_commissural | 0.151 | -0.295 | 0.087 |
| ILF_left | -0.071 | 0.268 | 0.072 |
| ILF_right | -0.222 | -0.41 | 0.168 |
| IOFF_left | 0.096 | -0.288 | 0.083 |
| IOFF_right | -0.262 | -0.327 | 0.107 |
| TF_left | 0.049 | -0.195 | 0.038 |
| TF_right | -0.076 | -0.559 | 0.312 |
| UF_left | 0.211 | -0.508 | 0.258 |
| UF_right | -0.833 | -0.834 | 0.696 |
| Y | |||
| attn_vig | -0.144 | -0.548 | 0.301 |
| problem_solving | 0.235 | -0.337 | 0.114 |
| process_speed | -0.758 | -0.806 | 0.650 |
| verbal_learning | -0.085 | -0.624 | 0.389 |
| visual_learning | 0.241 | -0.489 | 0.239 |
| work_memory | 0.269 | -0.301 | 0.090 |
| EA | -0.205 | -0.545 | 0.297 |
| ER_40 | -0.070 | -0.386 | 0.149 |
| RAD | -0.266 | -0.698 | 0.487 |
| RMET | 0.111 | -0.372 | 0.139 |
| TASIT_1 | -0.067 | -0.557 | 0.310 |
| TASIT_2_paradoxicalSarcasm | -0.149 | -0.608 | 0.370 |
| TASIT_2_simpleSarcasm | -0.041 | -0.538 | 0.290 |
| TASIT_2_sincere | 0.030 | -0.062 | 0.004 |
| TASIT_3_lies | 0.095 | -0.195 | 0.038 |
| TASIT_3_sarcasm | -0.211 | -0.582 | 0.339 |
| Note: | |||
|
\(r_s\) values > .45 are emphasized, following convention; \(\beta\) = standardized canonical function coefficient; \(r_s\) = structure coefficient; \(r_s^2\) = squared structure coefficient, here also communality \(h^2\) |
|||
comparison table
This table indicates if a given variable was found to be important (\(r_s\) ≥ .45) to its variate in neither sample, the GE sample only, the PRISMA sample only, or both samples.
Prediction
Participant scores on significant canonical variates can serve as new variables in subsequent analyses. We opted to see if participant scores predicted BSFS (social functioning) scores, given noted contributions of white matter, neurocognition, and social cognition to social functioning. Both simple and multiple linear regression models predicting BSFS scores were highly significant in both samples, with moderate-to-large correlation coefficients (\(R\)) and coefficients of determination (\(R^2\)).
GE
Multiple regression: Adjusted \(R^2\) = 0.155, model p = 0.
PRISMA
Multiple regression: Adjusted \(R^2\) = 0.332, model p = 0.
Classification
Lastly, we performed k-means clustering on participant scores, imposing a cluster size of two. The reason for this was to determine if the two clusters would be differentiated by categorical diagnosis, i.e., HC vs. SSD. We found high but imperfect overlap between diagnostic label and classification based on \(X\) and \(Y\) CV1 scores. The ARI was higher for the PRISMA than GE. We take this as support for the RDoC approach.
GE
PRISMA
