OPENNEURO:ds002778

we also calculated these measures using a bipolar scheme (C3-CP1 and C4-CP2) and found very similar results (p < 0.004, FDR corrected for each of the three measures examined)

Results, Waveform metrics differentiate PD medication state, paragraph 1

What the paper reports

• Wilcoxon signed rank, p < 0.004, FDR, over the three measures-corrected (n = 15 patients, on and off)

With C3-CP1 and C4-CP2, off is higher than on for sharpness ratio (W = 15, p = 0.0106, 13 of 15), steepness ratio (W = 4, p = 0.0015, 13 of 15) and PAC (W = 12, p = 0.0064 Methods' filters; W = 11, p = 0.0054 code's). FDR corrected over the three, 0.0106, 0.0044 and 0.0096 (0.0081), all below 0.05. None is below the paper's bound of 0.004 after correction (steepness 0.0044 is closest); uncorrected, only steepness is (0.0015).

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion (as for fig2-sharpness-off-on).

none of these measures differed between patients on and off medication over these electrodes ... sharpness ratio, p = 0.73, uncorrected, steepness ratio, p = 0.394, uncorrected PAC, p = 0.191, uncorrected

Results, Waveform metrics differentiate PD medication state, paragraph 2

What the paper reports

• sharpness ratio: Wilcoxon signed rank, p = 0.73, none-corrected (n = 15 patients, on and off)
• steepness ratio: Wilcoxon signed rank, p = 0.394, none-corrected (n = 15 patients, on and off)
• PAC: Wilcoxon signed rank, p = 0.191, none-corrected (n = 15 patients, on and off)

At the F7 and F8 composite, sharpness ratio p = 0.69 (d 0.31), steepness ratio p = 0.17 (d 0.57), PAC p = 0.0995 (Methods' filters) and 0.078 (code's), d 0.67; none below 0.05 uncorrected. Paper p = 0.73, 0.394 and 0.191.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion. The paper's argument is that these electrodes carry muscle artifact; here artifacts were not removed at any electrode.

Results were similar for T7 and T8, also near temporalis muscles (p > 0.05, uncorrected for all comparisons)

Results, Waveform metrics differentiate PD medication state, paragraph 2

What the paper reports

• Wilcoxon signed rank, p > 0.05, none-corrected (n = 15 patients, on and off)

At the T7 and T8 composite, sharpness ratio p = 0.36 (d 0.42), steepness ratio p = 0.95 (d -0.04), PAC p = 0.11 under both filter readings (d 0.56 and 0.59).

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.

There was no significant difference between these groups for beta waveform sharpness ratio (p = 0.363, uncorrected)

Results, Waveform metrics differentiate PD medication state, paragraph 3; Table 2

What the paper reports

• Wilcoxon rank sum, rank sum = 0.91, p = 0.3631, none (FDR corrected p = 1)-corrected (n = 15 patients off medication and 16 controls)

Rank sum z = 0.47, p = 0.64 (paper z = 0.91, p = 0.363); no difference. d = 0.09 is outside the expected 0.19 +/- 0.019; the claim of absence is judged on the test.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.

Somewhat surprisingly there was also no significant difference between PD patients off medication and control participants for the PAC calculation (p = 0.236, uncorrected)

Results, Waveform metrics differentiate PD medication state, paragraph 3; Table 2

What the paper reports

• Wilcoxon rank sum, rank sum = 1.19, p = 0.236, none-corrected (n = 15 patients off medication and 16 controls)

Rank sum z = 0.99, p = 0.32 with the Methods' filters; z = 1.07, p = 0.29 with the code's (paper z = 1.19, p = 0.236). d = 0.20.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.

these changes were in the context of significant clinical improvement with medication as measured by Unified Parkinson's Disease Rating Scale measures (Wilcoxon signed rank test statistic = 3.5, p = 0.003)

Results, Waveform metrics differentiate PD medication state, paragraph 1

What the paper reports

• Wilcoxon signed rank, signed rank = 3.5, p = 0.003 (n = 15 patients, on and off)

The dataset's "Total UPDRS" (16 to 58; it exceeds "UPDRS 18-26" by 1 to 14, which fits the motor examination, items 18 to 31) is lower on medication in 12 patients, equal in 2, higher in 1. Signed rank W = 3.5, p = 0.0033 (scipy's default drops the two ties), the paper's values. The dataset holds no other part of the UPDRS, so both readings of the criterion give this field. Items 18 to 26 alone give W = 6.5, p = 0.011.

How the test differed from the paper’s method

• The UPDRS fields have no description in the dataset; "Total UPDRS" is taken as the motor total from its values (it exceeds "UPDRS 18-26" by 1 to 14), which is consistent with the values, not shown.

In general, sensorimotor electrode oscillations had a sharper peak with a steeper decay, falling in quadrant 4 (Q4) in Figure 4

Results, Waveform shape, paragraph 1; Figure 4A, C

Of the 60 patient points (C3 and C4, off and on), 48 (80%) have positive log10 peak-to-trough and negative log10 rise-to-fall ratio (Q4); Q1 6, Q2 4, Q3 2. Q4 holds 13, 10, 13 and 12 of the 15 points at C3 off, C3 on, C4 off and C4 on.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.

In contrast, electrodes closest to temporalis muscles had a more variable waveform shape pattern

Results, Waveform shape, paragraph 1; Figure 4B

Reading (a), the most common quadrant holds 68% of the 60 F7 and F8 points (41, Q2) against 80% (48, Q4) at C3 and C4; reading (b), RMS distance to the centroid 0.093 against 0.076. Both readings agree. The F7 and F8 points concentrate in Q2 (sharper troughs, steeper rises).

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.

Although not shown, control participant data exhibited the same pattern (i.e., sensorimotor electrode data fell mainly in Q4)

Results, Waveform shape, paragraph 1

25 of the 32 control points (78%) are in Q4 (13 of 16 at C3, 12 of 16 at C4); Q1 2, Q2 2, Q3 3.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.

Medication reduced beta waveform sharpness ratio (p = 0.012, FDR corrected; Fig. 2)

Results, Waveform metrics differentiate PD medication state, paragraph 1; Figure 2A, B; Table 1

What the paper reports

• Wilcoxon signed rank, signed rank = 12, p = 0.006, none (FDR corrected p = 0.0121)-corrected (n = 15 patients, on and off)
• Wilcoxon signed rank, signed rank = 12, p = 0.0121, FDR-corrected (n = 15 patients, on and off)

Composite C3 and C4 beta sharpness ratio is higher off than on in 13 of 15 patients, signed rank W = 13 (expected 12 +/- 1.2), p = 0.0076 (exact 0.0054), FDR corrected over the six tests of Table 1 p = 0.0228 under all 12 combinations of filter readings. Cohen's d = 0.73, below 0.86 +/- 0.086. Without sub-pd6 and sub-pd16, whose on sessions come from preprocessed files, W = 11, p = 0.016, d = 0.81.

How the test differed from the paper’s method

• No samples rejected and no electrode excluded, as the dataset holds neither the paper's manual artifact marks nor its excluded electrodes (claims file readings (b); reading (a), the dataset's own marks, has nothing to apply since every channels.tsv says good and the events files hold only a start trigger).
• The paper computed its measures on preprocessed MATLAB files that are not public; here the preprocessing of the Methods is applied to the BDF files.

steepness ratio (p = 0.012, FDR corrected; Fig. 2) in PD patients over sensorimotor electrodes

Results, Waveform metrics differentiate PD medication state, paragraph 1; Figure 2A, B; Table 1

What the paper reports

• Wilcoxon signed rank, signed rank = 9, p = 0.004, none (FDR corrected p = 0.0121)-corrected (n = 15 patients, on and off)
• Wilcoxon signed rank, signed rank = 9, p = 0.0121, FDR-corrected (n = 15 patients, on and off)

Beta steepness ratio is higher off than on in 12 of 15 patients, p = 0.0125 (exact 0.0103), FDR corrected 0.0249; d = 0.70 is within 0.68 +/- 0.068 but W = 16 is outside 9 +/- 0.9. Without sub-pd6 and sub-pd16, W = 5, p = 0.0046, d = 0.75.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion (as for fig2-sharpness-off-on).

PAC also decreased with medication (p = 0.017, FDR corrected; Fig. 2), as has been previously published

Results, Waveform metrics differentiate PD medication state, paragraph 1; Figure 2A, B; Table 1

What the paper reports

• Wilcoxon signed rank, signed rank = 15, p = 0.011 (n = 15 patients, on and off)
• Wilcoxon signed rank, signed rank = 15, p = 0.0166, FDR-corrected (n = 15 patients, on and off)

Beta PAC is higher off than on in 13 of 15 patients; W = 13, p = 0.0076 with the Methods' filters (118 and 123 samples) and W = 12, p = 0.0064 with the code's (231 and 240 taps); FDR corrected 0.0228. d on raw PAC is 0.85 and 0.86, within 0.83 +/- 0.083; on log10 PAC 0.73 and 0.86. The statistic is outside 15 +/- 1.5 under both filter readings. The criterion's last sentence names only the test and d; taken alone it would make the claim reproduced, but the statistic is an expected value of the claim and is out of tolerance, so the outcome is partially reproduced; which sentence governs is asked of a reader (questions/openneuro-ds002778.yaml). Dropping the filtered edges, as pacpy's firf does, changes PAC by at most 5.1% in any recording (ratio 0.949 to 1.043) and leaves W at 13 and 12. Without sub-pd6 and sub-pd16, W = 13 and 12, p = 0.023 and 0.019, d = 0.73 and 0.75.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion (as for fig2-sharpness-off-on).
• The criterion combines "As for fig2-sharpness-off-on" (statistic and d within tolerance) with a last sentence naming the test and d only; the expected statistic was kept as part of the judgement.

found that topographies were consistent with a sensorimotor origin (Fig. 2C)

Results, Waveform metrics differentiate PD medication state, paragraph 2; Figure 2C legend

Signed rank tests at each of the 32 electrodes. Steepness ratio meets the criterion (smallest p at C4, 0.023; F7, F8, T7, T8 p 0.21 to 0.57). PAC with the Methods' filters has its smallest p at CP6 (0.027), and F7 has p = 0.041, so it meets the criterion under the FDR reading only; with the code's filters Fz and CP6 tie for the smallest p (0.020), and Fz is outside the sensorimotor set. Sharpness ratio has its smallest p at P8 (0.0064), then CP6 (0.017) and AF3 (0.031); C3 is fourth (0.0995). No electrode survives FDR over 32 for any measure. Two of three measures meet the criterion under one reading, one does not under any.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.
• The paper's topographies were drawn with EEGLAB topoplot from test statistics; here the test at each electrode is listed and plotted as -log10 p, as the criterion uses only the tests.

We found these values to be overall similar to those we found for β, with the exception of PAC

Results, Waveform metrics differentiate PD medication state, paragraph 4; Table 1

What the paper reports

• alpha sharpness ratio: Wilcoxon signed rank, signed rank = 20, p = 0.0278, FDR-corrected (n = 15 patients, on and off)
• alpha steepness ratio: Wilcoxon signed rank, signed rank = 12, p = 0.0121, FDR-corrected (n = 15 patients, on and off)

With the 192-tap filter, alpha sharpness ratio W = 18, p = 0.0171, d = 0.71 (11 of 15 higher off) and alpha steepness ratio W = 21, p = 0.0268, d = 0.58 (12 of 15); with 118 samples, W = 19, d = 0.66 and W = 21, d = 0.59. FDR corrected p 0.0256 to 0.0321. Both tests hold and three of four values are within tolerance under both readings; the steepness statistic (21) is outside 12 +/- 1.2.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.

for PAC, the difference between off medication and on medication in β was significantly larger than the difference in μ/α for the same comparison (Wilcoxon signed rank test statistic = 11, p = 0.005)

Results, Waveform metrics differentiate PD medication state, paragraph 4

What the paper reports

• Wilcoxon signed rank, signed rank = 11, p = 0.005 (n = 15 patients)

On raw PAC the beta off minus on difference exceeds the alpha one in 12 or 13 of 15 patients, W = 17 (18 for the code's beta filters with reading c), p = 0.015 to 0.017, under all six combinations of filter readings. W is outside 11 +/- 1.1. On log10 PAC (variant) W = 40 to 54, p = 0.26 to 0.73. The criterion has no partial clause; partially reproduced follows the protocol's definition (the effect is present but weaker), and the question is asked of a reader (questions/openneuro-ds002778.yaml).

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.

sharpness ratio and PAC were correlated in both on and off medication states in sensorimotor electrodes (off medication ... r = 0.70, p = 0.004; on medication ... r = 0.74, p = 0.002; Fig. 3)

Results, Relationships between sharpness/steepness ratio and PAC, paragraph 1; Figure 3

What the paper reports

• off medication: Spearman correlation, r = 0.70, p = 0.004 (n = 15 patients)
• on medication: Spearman correlation, r = 0.74, p = 0.002 (n = 15 patients)

Spearman r between composite beta sharpness ratio and PAC, Methods' filters, off 0.59 (p = 0.022, below 0.63 to 0.77) and on 0.79 (p = 0.0005, within); code's filters off 0.63 (p = 0.012, within) and on 0.84 (p = 0.0001, above 0.666 to 0.814). Both positive and significant under both readings, with one r out of tolerance in each. Without sub-pd6 and sub-pd16, off r = 0.51 (p = 0.074) and 0.57 (p = 0.041), on r = 0.87 and 0.88.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.

steepness ratio and PAC are also highly correlated for both medication states (off medication ... r = 0.74, p = 0.002; on medication ... r = 0.68, p = 0.006)

Results, Relationships between sharpness/steepness ratio and PAC, paragraph 1; Figure 3

What the paper reports

• off medication: Spearman correlation, r = 0.74, p = 0.002 (n = 15 patients)
• on medication: Spearman correlation, r = 0.68, p = 0.006 (n = 15 patients)

Methods' filters, off r = 0.80 (p = 0.0003, within 0.666 to 0.814), on r = 0.45 (p = 0.095, not significant); code's filters, off r = 0.80 (p = 0.0004), on r = 0.53 (p = 0.041, below 0.612 to 0.748). Under the code's filters both are positive and significant with one r out of tolerance. Without sub-pd6 and sub-pd16 (a variant) both r are within tolerance under both filter readings (off 0.75, on 0.64 and 0.65, p 0.0033 to 0.019), so the on medication miss rests on these two patients.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.

Effect size increased for all measures, at least marginally, when using the Q4 shape as an inclusion criterion (Table 3)

Results, Waveform shape, paragraph 2; Table 3

With the 13 patients who have at least one electrode in Q4 off and on, d rises for all three measures under every reading; sharpness 0.73 to 0.99 (within 0.95 +/- 0.095), steepness 0.70 to 0.82 (above 0.70 +/- 0.07), raw PAC with the Methods' filters 0.85 to 0.95 (above 0.83 +/- 0.083; code's 0.86 to 0.96, log10 0.73 to 1.06 and 0.86 to 1.07). The all-patient PAC d of Table 3 (0.75) is matched by log10 PAC with the Methods' filters (0.73), not by raw PAC (0.85, given as measured, first reading). The Q4 group has 13 patients where the paper has 9 (q4-patient-counts, not reproduced), so the Q4 values come from a different and larger subset than Table 3's. Without sub-pd6 and sub-pd16 (12 Q4 patients) d rises for sharpness (0.81 to 1.09) and raw PAC (0.73 to 0.87) and falls slightly for steepness (0.75 to 0.72).

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.

however steepness ratio did distinguish patients from controls (p = 0.033 using the average referencing scheme and p = 0.028 for a bipolar referencing scheme, both FDR corrected)

Results, Waveform metrics differentiate PD medication state, paragraph 3; Table 2

What the paper reports

• average reference: Wilcoxon rank sum, rank sum = 2.53, p = 0.011, none (FDR corrected p = 0.0331)-corrected (n = 15 patients off medication and 16 controls)
• average reference: Wilcoxon rank sum, rank sum = 2.53, p = 0.0331, FDR-corrected (n = 15 patients off medication and 16 controls)
• bipolar montage: Wilcoxon rank sum, p = 0.028, FDR-corrected (n = 15 patients off medication and 16 controls)

With the average reference, beta steepness ratio is higher off than in controls, z = 2.37 (within 2.53 +/- 0.253), d = 0.85 (within 0.88 +/- 0.088), p = 0.0177 uncorrected, but FDR corrected over the six tests of Table 2 p = 0.0531 under every filter reading, just above 0.05 (the alpha steepness test has p = 0.0143, so the two smallest of six give 0.0531). With the bipolar pairs z = 2.37, p = 0.0177 as a single test (significant), 0.0531 FDR corrected over the three bipolar measures. Both outcomes the criterion names other than not reproduced need the corrected average reference test to hold, and it does not, so the claim is not reproduced (changed from partially reproduced after verification round 1). The miss rests on the uncorrected p (0.0177 against the paper's 0.011; BH over six needs p < 0.0167 here). Holm 0.0885, Bonferroni 0.106 and Benjamini-Yekutieli 0.130 over six are further from 0.05; BH over the three beta tests gives 0.0531. Without sub-pd6 and sub-pd16 z = 2.15, p = 0.032 uncorrected.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.

When we tested the CSD referencing scheme using the current method for PAC calculation, there was a significant difference between patients off medication and controls (p = 0.022)

Results, Waveform metrics differentiate PD medication state, paragraph 3

What the paper reports

• Wilcoxon rank sum, p = 0.022 (n = 15 patients off medication and 16 controls)

With the CSD transform, beta PAC at C3 and C4 is higher off (0.032) than in controls (0.025), z = 1.78, p = 0.075 under both filter readings, d = 0.35, against the paper's p = 0.022. The CSD montage increases the off against on difference (d = 1.20, p = 0.0106); every CSD variant gives a higher mean off than on, with 12 or 13 of 15 patients higher off (p 0.0045 to 0.015). The result depends on the spline parameters and no variant reaches 0.05 off against controls; m = 4 gives p = 0.30 and 0.29, lambda 1e-4 gives 0.18 and 0.21, lambda 1e-6 gives 0.11 and 0.12 (Methods' and code's filters). The CSD raises 50 to 150 Hz power relative to beta at C3 and C4 (median ratio 0.22 in controls and 0.34 off, against 0.12 and 0.18 with the average reference). Whether MNE's output equals the CSD toolbox's was not checked.

How the test differed from the paper’s method

• CSD computed with MNE's spherical spline (stiffness m = 3, lambda 1e-5, 50 Legendre terms, the CSD toolbox's defaults) on the BioSemi 32 template positions, as the dataset holds no digitized positions; the paper used the MATLAB CSD toolbox (Methods, Data preprocessing).
• No artifact rejection or electrode exclusion.

Plots confirm that there is a β oscillation as well as a prominent α/μ oscillation

Figure 1C legend; Results, Waveform metrics differentiate PD medication state, paragraph 4

Welch spectra (Hamming 256, overlap 128, nFFT 450) averaged over C3 and C4 and over the participants of each group. Controls have local maxima at 10.24 Hz and 18.2 Hz. The off group's mean spectrum has its only maximum from 6 to 13 Hz at 7.96 Hz, below the 8 to 12 Hz band, and none from 13 to 30 Hz; the on group's has none. Over all electrodes off has beta maxima at 17.1 and 21.6 Hz and no alpha maximum. The on mean is dominated by sub-pd16 on (about 100 times the power of others), but without it there is still no maximum. With the mean of log power, off has an alpha maximum at 9.1 Hz and no beta maximum. Single recordings have an alpha maximum in 8 of 15 off, 8 of 15 on and 11 of 16 controls, and a beta maximum in 9, 10 and 14.

How the test differed from the paper’s method

• Spectra computed without artifact rejection; artifacts weigh on a mean of raw power more than on the ratio measures.

with the exception of PAC, which was non-significant for μ/α, while being significant for β

Results, Waveform metrics differentiate PD medication state, paragraph 4; Table 1

What the paper reports

• Wilcoxon signed rank, signed rank = 30, p = 0.0881, FDR-corrected (n = 15 patients, on and off)

FDR corrected over Table 1, alpha PAC off against on gives p = 0.0609 with a 192-tap phase filter (reading a, W = 27), 0.0783 with the code's 231 and 240 taps (reading b, W = 29), and 0.0356 with 118 samples (reading c, W = 23). Readings a and b find no difference, reading c does. The criterion takes the outcome all readings give; the null holds under two of three, so the claim is not reproduced (changed after verification round 1 from reproduced under the assumption of a 192-tap filter). d on raw PAC is 0.70, 0.72 and 0.66, above 0.51 +/- 0.051 under every reading; on log10 PAC 0.84, 0.82, 0.70. Without sub-pd6 and sub-pd16 (a variant) the uncorrected p is 0.064 to 0.087 and the null holds after FDR correction under all three readings. The question of the alpha phase filter length is for the paper's authors, since the Methods give 231 ms for the beta filter only, and the code holds no alpha analysis.

How the test differed from the paper’s method

• The alpha phase filter length is not given; the three readings of the claims file (192, 231 and 118 taps) are run and disagree.
• No artifact rejection or electrode exclusion.

Not quoted: Values from Table 2; the text says only that alpha values were overall similar to beta, except PAC.

What the paper reports

• alpha sharpness ratio: Wilcoxon rank sum, rank sum = 1.03, p = 1, FDR-corrected (n = 15 patients off medication and 16 controls)
• alpha steepness ratio: Wilcoxon rank sum, rank sum = 2.53, p = 0.0331, FDR-corrected (n = 15 patients off medication and 16 controls)
• alpha PAC: Wilcoxon rank sum, rank sum = 2.17, p = 0.0602, FDR-corrected (n = 15 patients off medication and 16 controls)

Alpha steepness ratio, off against controls, z = 2.45 (192 taps) or 2.53 (118 samples), within 2.53 +/- 0.253, d = 0.89, within 0.87 +/- 0.087, p = 0.0143 or 0.0114 uncorrected, FDR corrected 0.0531 under every reading, just above 0.05. Sharpness ratio (corrected 0.57 or 0.72) and PAC (0.12 to 0.41) do not differ after correction, as the claim says. Both outcomes the criterion names other than not reproduced need the steepness result to hold, so the claim is not reproduced (changed from partially reproduced after verification round 1). Holm, Bonferroni and Benjamini-Yekutieli over six give 0.086 to 0.130; BH over the three alpha tests, not the family the criterion names, gives 0.043.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion.

Out of 15 patients, nine patients met this criterion for C3 and eight met the criterion for C4 ... Using this approach there were nine total Q4 observations for each waveform metric

Results, Waveform shape, paragraph 2

Patients with both off and on in Q4 are 10 at C3, 10 at C4, 7 at both and 13 at one or more, against 9, 8, 8 and 9 (+/- 1). The C3 and both counts are within tolerance, the C4 count and the at-least-one count are not, in either electrode order (swapped, C3 10 against 8). Without sub-pd6 and sub-pd16 the counts are 9, 9, 6 and 12. Of the 60 points, 16 lie within 0.01 of an axis (log10 ratios), and 10 of them decide membership at their electrode (moving their near-axis coordinates across zero would change whether the patient counts there), in pd3, pd5, pd11, pd12, pd22, pd26 and pd28; the per-patient values are printed.

How the test differed from the paper’s method

• No artifact rejection or electrode exclusion; the selection depends on the shape measures, which the paper computed after rejection.