Correlating evoked electromyography and anatomic factors during SCS implantation with short-term outcomes

Abstract

Introduction:

We examine how intraoperative neuromonitoring (IONM) during spinal cord stimulation (SCS) varies between a high resolution investigational SCS paddle (HR-SCS) paddle and commercial paddle is explored. Further the presence of evoked motor responses (e.g. electromyography (EMG) in painful regions during surgery is correlate to outcomes.

Materials and Methods:

We utilized HR-SCS to assess EMG response from 18 patients (NCT05459324). Maximum percentage change in root mean squared (maxRMS) EMG values were determined. Correlations were performed with MRI measurements and patient outcomes collected pre-operatively and at 3 months (Numerical Rating Scale (NRS), McGill Pain (MPQ), Beck Depression Inventory (BDI), Oswestry Disability Index (ODI), and Pain Catastrophizing Score (PCS)).

Results:

Of the 18 patients (12F:6M; mean age 56; 8 with neuropathic pain, 8 PSPS, 2 CRPS), nine had a response at 3 months based on 50% reduction in NRS, 14 by achieving MCID on NRS and 11 by reaching MCID on ≥3 outcome metrics. The anterior posterior diameter (APD) of spinal column at level of testing correlated with all three responses (p<0.05). We examined RMS at muscles correlating with individual patient pain distributions and found correlations between RMS and MCID NRS and MCID ODI (p<0.05). MaxRMS in abductor hallucis correlated with improvement in NRS and ODI across the group (p<0.05).

Conclusion:

We found that eliciting EMGs over the painful areas during surgery resulted in improvement in pain intensity and disability. Obtaining stimulation of AH was more predictive of pain improvement than any other muscle group and APD alone correlated with improvements in pain intensity and holistic outcomes. These pilot data suggest that implanters should consider APD and EMG responses from painful regions and AH during surgery.

Introduction

Chronic pain is a pervasive condition affecting millions of individuals worldwide, significantly burdening patients and healthcare systems^1-4. Chronic pain often compromises the quality of life, particularly in those resistant to standard therapeutic interventions, leading to physical and psychological distress and resulting in economic consequences^3-5. Spinal cord stimulation (SCS) is an effective intervention for conditions such as persistent spinal pain syndrome (PSPS), neuropathic pain, and complex regional pain syndrome (CRPS)^6-10. SCS offers pain relief of varying degrees to many patients; however, there are some who do not achieve adequate relief in specific areas^{6, 7}. Such variability in outcomes underscores the need for more personalized and targeted stimulation techniques¹¹. Evoked motor responses have emerged as a potentially transformative tool for enhancing the precision of SCS targeting^{12, 13}. However, it remains unclear how targeting specific areas of the dorsal column results in pain relief in particular sites^{12, 14-16}.

Since the effectiveness of SCS therapy depends on correct placement to align with the regions of pain, there is a need for objective measures. Some have postulated that evoked motor response during IONM may provide a surrogate for dermatomal pain coverage. Recording EMG signals has been shown to accurately locate the physiological midline of the spinal cord during SCS placement, resulting in effective pain reduction in patients¹⁷. In a recent systematic review, the authors found that the use of IONM during SCS surgery results in better pain relief, fewer neurologic deficits, and shorter operation times¹⁸. We have also shown that dorsal cerebrospinal fluid (CSF) and CSF thickness affected amplitudes needed to obtain a motor response¹⁹.

Here we utilize a high resolution investigational SCS paddle (HR-SCS) to assess evoked motor responses from 18 patients (NCT05459324). In this study, we evaluate how patient specific factors, including root mean square (RMS) of signals in muscle groups affected by pain and anatomic factors, influence outcomes.

Materials and Methods

Patients were recruited to this IDE (FDA IDE code: G210346) study if they were suffering from ongoing leg and back pain, had a positive trial with SCS showing more than a 50% improvement on the numerical rating scale (NRS), had passed a psychological evaluation, and had undergone pre-surgery MRI imaging. In our practice, pain specialists performed a percutaneous trial and we recruited from patients who were subsequently referred for paddle placement. The study was registered on ClinicalTrials.gov (Identifier: NCT05459324). Every patient gave their written consent before participating in this IRB approved study (Albany Medical College IRB, #5151 and #6426).

Surgical workflow

Once patients provided consent, they completed baseline outcome measures for pain, disability, and depression. These assessments included the Numerical Rating Scale (NRS), Beck Depression Inventory (BDI), McGill Pain Questionnaire (MPQ), Oswestry Disability Index (ODI), Pain Catastrophizing Scale (PCS), and a drawing illustrating their areas of pain. All these questionnaires were completed in the clinic and/or with a study coordinator/medical team available for questions. Demographic information including sex, age, diagnosis, and duration of illness was documented (Table 1). Patients then underwent a standard of care laminotomy with intraoperative neuromonitoring (IONM) as previously described¹². Patients completed patient reported outcomes (PROs) again at 3 months.

Table 1: Subject Demographics.

M: Male. F: Female. -PSPS (Persistent Spinal Pain Syndrome). NP: Neuropathic pain. CRPS: Complex Regional Pain Syndrome. B/L: Bilateral. L: Left. R: Right. MCID: Minimal Clinically Important Difference. X: Denotes “yes” for responder. mA: Milliampere. APD: Anterior posterior diameter. IPD: Interpedicular distance. dCSF: Dorsal cerebrospinal fluid. mm: millimeter. Likert scale: 1 =no areas covered, 2 = low response of some painful areas, 3 = low response of all painful areas or majority of areas uncovered with low/mid response in those covered, 4 = coverage of some painful areas as a mid-responder, 5 = coverage of most areas as a mid-responder, 6 = majority of painful areas covered robustly as a mid-high responder 7 = all painful areas covered mid-high response ND: No data

Patient	Age	Sex	Diagnosis	Duration of Illness (years)	Pain Distribution	Responder at the 3m follow-up based on 50% NRS	Responder at the 3m follow-up based on MCID NRS	Responder at the 3m follow-up based on MCID 3 or more	Minimum Stimulation Amplitude (mA)	Maximum Stimulation Amplitude (mA)	Commerical paddle type	MRI-obtained measurement: APD (mm)	MRI-obtained measurement: IPD (mm)	MRI-obtained measurement: dCSF (mm)	Likert Scores
2	60	M	PSPS	8	B/L - low back		X	X	5	7.5	Medtronic Specify™ 5-6-5 Lead	15.8	20	4	7: complete coverage of the back
3	46	M	NP	5	B/L - low back, glutes, hamstrings, calves	X	X	X	1	2	Boston Scientific CoverEdge™ 32	11.3	14.8	2.5	3: coverage of the back and right calf, no glute or hamstring coverage
4	29	M	PSPS	4	R - glutes, hamstring, calf		X	X	10	10	Nevro Surpass™ Surgical Lead	15.2	19.5	5	2: coverage of right low back, no right leg coverage
5	43	F	CRPS	4	B/L - full back, low back (above glutes), glutes, posterior shoulder. L - calf, knee, shin. R - knee	X	X	X	7.5	10	Nevro Surpass™ Surgical Lead	14.8	19.6	5	4: coverage of the low back and calf, no glute coverage
6	40	F	NP	18	L - low back (above glute), glute, hamstring, calf	X	X		5	10	Medtronic Specify™ 5-6-5 Lead	17	21.5	4.8	6: coverage of the low back, left glute, and hamstring, minimal calf coverage
7	52	F	NP	10	Upper and lower back				10	10	Medtronic Specify™ 5-6-5 Lead	17.6	18.8	3.9	2: minimal low back coverage
8	67	M	NP	15	B/L - low back. R - knee		X		10	10	Abbott Penta™ Paddle Lead	19	19.3	5.3	5: coverage of low back, no knee coverage
9	57	F	NP	7	B/L - low back (above glute). L - glute, hamstring, calf				10	10	Boston Scientific CoverEdge™ 32	18.1	20	5.4	1: no coverage
12	78	F	NP	7	B/L - low back and abdominals				3	10	Boston Scientific CoverEdge™ 32	18	21.5	4.7	5: coverage of low back
21	57	F	CRPS	>4	B/L - full back, low back (above glutes), posterior leg (glutes, calf, hamstring)	X	X	X	7.5	10	Boston Scientific CoverEdge™ 32	ND	ND	ND	4: minimal coverage of back and left leg
22	60	F	NP	19	B/L - low back (above glutes), anterior and posterior leg (quads, calf, hamstrings, shins)	X	X	X	10	10	Boston Scientific CoverEdge™ 32	15	17.6	4.3	4: minimal coverage of back
23	67	F	PSPS	12	B/L - low back (above glutes). R - quad and hamstring	X	X	X	2.5	10	Boston Scientific CoverEdge™ 32	14.1	17.3	3.1	5: good coverage of the left side
24	51	F	PSPS	5	B/L - low back, glutes, quads and hamstrings		X		10	10	Boston Scientific CoverEdge™ 32	14.4	17.5	4.8	5: coverage of the back, no right leg coverage
25	78	F	PSPS	1.5	B/L - glutes and hamstrings				5	9	Nevro Surpass™ Surgical Lead	15.2	19.3	4.9	3: only right quad coverage, no left or right hamstring coverage
26	75	M	PSPS	>20	B/L - low back (above glutes). R - pelvis	X	X	X	5.5	7.5	Nevro Surpass™ Surgical Lead	12.2	18	4.7	6: low to mid coverage of back
27	34	F	PSPS	>4	B/L - mid back (around spine), low back. L - hamstring, calf, knee. R - posterior shoulder, hip, hamstring		X	X	4.5	7	Nevro Surpass™ Surgical Lead	14.4	17.2	3.6	7: coverage of all muscle groups involved at high levels
28	71	M	NP	20	B/L - upper and lower back, posterior neck, dorsal arms (shoulders to fingertips), glutes, hamstrings, calves	X	X	X	10	10	Boston Scientific CoverEdge™ 32	12.3	21.4	5.6	4: some coverage of low back and right calf, no hamstrings, gluts, or left calf coverage
29	49	F	PSPS	15	B/L - glutes, quads, anterior lower leg	X	X	X	10	10	Nevro Surpass™ Surgical Lead	14.8	18.7	6.2	4: coverage of quads, front of leg, no coverage of gluts

Pre-operative MRI images were obtained for every patient. The pre-operative axial spinal MRI of each patient was assessed to measure the CSF area at the level of the stimulated contacts. CSF thickness was then measured using a tracing tool available on the iSite radiology program (Philips Healthcare Informatics, Foster City, CA, USA), as described previously. Anterior posterior diameter (APD) was measured on the axial plane, from the middle of the posterior vertebral body to the anterior aspect of the lamina, at the level of relevant contacts.

Compound motor action potentials, evoked motor responses (termed ‘EMG’ for remainder of manuscript), and somatosensory evoked potentials (SSEPs) were used during surgery as standard of care. In the study, we see if EMG responses differ with the experimental lead and commercial lead. We use IONM so that we can perform the cases under general anesthesia and ensure laterality and patient safety with paddle leads²⁰. A cascade PRO IONM system (Cadwell Inc., Kennewick WA) was utilized to monitor and record EMG data. Testing was conducted on nine muscle groups bilaterally. From distal to proximal, we assessed the abductor hallucis (AH), tibialis anterior (TA), medial gastrocnemius (MG), quadriceps (QUAD), bicep femoris (BF), adductor magnus (ADD), gluteus maximus (GLUT) and lower and upper rectus abdominis (LAB; UAB) (Figure 1). Following standard of care laminotomy, HR-SCS ( Micro-Leads Inc., Somerville, MA) was first placed and EMGs were monitored when the device was stimulated. Following testing, it was removed and the commercial SCS paddle (model type and company based on lead used in trial) was placed, EMG responses documented, and implanted. Placement of paddles was confirmed with C-arm fluoroscopic imaging. We aligned commercial and HR-SCS leads based on anatomical placement on these images. In 16 out of 18 patients, anatomical midline overlapped with the physiological midline. In the 2 remaining patients, the evoked responses observed in the muscles on the right body were smaller compared to responses on the left body. In these cases, physiological midline was inconclusive.

Figure 1:

Nine muscle groups whose data were recorded to investigate their responses at applied stimulation. AH- Abductor Hallucis, TA- Tibialis Anterior, MG- Medial Gastrocnemius, BF- Biceps Femoris, ADD- Adductor Magnus, QUAD- Quadriceps, GLUT- Gluteus, LAB- Lower Abdominis, UAB- Upper Abdominis. Thoracic spinal levels T6-T10 levels where HR-SCS paddles were placed to investigate muscle responses are shown (red lightning bolt).

A longitudinal tripole (+/−/+) at both the top and bottom of the commercial and HR-SCS paddles were tested in 9 patients. In the initial 9 patients, the HR-SCS only allowed for testing of the bottom tripole. Stimulation was administered at a frequency of 60Hz with a pulse duration of 300μs through a hand-held programmer. Amplitudes were progressively raised by a 0.5mA step size until reaching either a motor threshold or maximum threshold (10mA). The minimum stimulation amplitude was defined as motor threshold based on EMG changes. Maximal stimulation amplitude was the highest amplitude used for stimulation in the OR.

Signal Processing

All EMG signals were processed offline in MATLAB R2022b (MathWorks, Natick, MA, USA). EMG signals were denoised using a previously developed algorithm, which involved applying a 10%-threshold and removing values exceeding the threshold¹³. The RMS of signals were computed at each amplitude (stimulation-ON) for all patients and normalized with respect to baseline (stimulation-OFF) as previously described²¹. At the individual level, maximum percentage change in RMS (maxRMS) was determined for amplitudes ≤ 10mA across all 8 mediolateral contacts.

Individual patient motor heatmaps were created in GraphPad Prism-10 (GraphPad Software, Inc., San Diego, California, USA) for all nine muscle groups at the corresponding spine level(s) (Figure 2). Based on maxRMS, patients were then divided into responders (values ≥50% RMS change with respect to baseline) and non-responders (values <50% RMS change with respect to baseline). The responder group was further divided into low (51-100 %RMS change), mid (101-500 %RMS change), and high (501-max %RMS change) responders.

Figure 2:

A characteristic motor recruitment heat map was created for each muscle. Here we show an example for Tibialis Anterior (TA) at 0.5-10mA (A, B). (A) depicts HR-SCS results while (B) depicts commercial paddles. Responders were determined based on values >50% RMS change with respect to baseline and non-responders were classified as values <50%. Responders were separated into low (51-100%RMS change), mid (101-500%), and high (501-max%) responders. These values were decided based on the change in percentage threshold.

The patient heatmaps were scored with a Likert scale to assess whether evoked motor response correlated with the patient’s individualized pain pattern. Specifically, on a 7 point scale: 1 (no areas covered), 2 (low response of some painful areas), 3 (low response of all painful areas or majority of areas uncovered with low/mid response in those covered), 4 (coverage of some painful areas as a mid-responder), 5 (coverage of most areas as a mid-responder), 6 (majority of painful areas covered robustly as a mid-high responder) to 7 (all painful areas covered mid-high response) (Figure 3). Robustness of coverage was based on %RMS change.

Figure 3.

Likert Scale Determination. Likert scale scores were determined to assess the likelihood of pain relief based on extent of coverage. This 7-point Likert scale was scored as follows based on degree of response on heatmaps for each muscle: 1 (no areas covered), 2 (low response of some painful areas), 3 (low response of all painful areas or majority of areas uncovered with low/mid response in those covered), 4 (coverage of some painful areas as a mid-responder), 5 (coverage of most areas as a mid-responder), 6 (majority of painful areas covered robustly as a mid-high responder) to 7 (all painful areas covered mid-high response). Robustness of coverage was based on %RMS change. The locations of pre-operative pain reported by individual subjects are depicted by black shaded areas on the human diagram. (A) represents a subject with a Likert score of 1 (B) represents a subject with a Likert score of 4 and (C) represents a subject with a Likert score of 7.

Statistical Analysis

Statistical analyses were performed in GraphPad Prism-10 (GraphPad Software, Inc., San Diego, California, USA). The difference in PRO scores between baseline and 3 months was documented. Percent change (Mean±SEM) and raw score change between the 3-month postoperative score and the preoperative baseline score were calculated. Previously established minimal clinically important difference (MCID) values for SCS treatment of pain, across all five outcome measures, were utilized. An improvement of at least 2.0 points on the NRS, 6.9 points on the BDI, 1 point on the MPQ, 8.2 points on the ODI, and 1.9 points on the PCS is clinically meaningful to the patient^22-25. Responders were those that achieved 50% pain relief on NRS. We termed those with MCID on NRS as MCID NRS responders and those with MCID on 3 or more outcome metrics as MCID >3 responders. Pearson’s correlation analysis was performed to determine the correlation values between pain outcome measures and MRI parameters. The D’agostino& Pearson’s test was used for testing normality. Some patients did not complete all patient reported outcomes and any missing data was omitted and not included in the analysis. Outliers were excluded and defined as +/− greater than 2 standard deviations away from the mean.

Results

The mean (SEM) percent change improvement in NRS score (n=18) was −45.6 + 7.7%, MPQ (n= 15) was −6.6% + 26.4%, ODI (n=15) was −30.2% + 8.3%, PCS (n=14) was −41.8% + 10.5%, and BDI (n=15) was −15.4% + 17.4% (Table 2). Nine of 18 patients were deemed responders based on 50% improvement in NRS at 3 months. Fourteen of 18 met MCID for NRS. Eleven of 15 met MCID for MPQ, 10 of 15 on MCID for ODI, 11 of 14 on MCID for PCS, and 7 of 15 on MCID for BDI (Table S1). Eleven of 18 patients achieved MCID on three or more outcome metrics. We examined factors which correlated with whether patients were deemed a responder. Percentage change in NRS inversely correlated with duration of illness (r= −0.50, p=0.03), with no other demographic correlations observed. The MRI measurement of APD correlated with response on all three primary endpoints (r=0.52, p=0.03 (% change NRS); r=0.52, p=0.03 (MCID on NRS); r=0.71, p=0.001 (MCID on 3 outcomes)).

Table 2:

Average raw score change and percent change for each pain outcome at the 3-month follow-up.

Outcome Measure	Average Raw Score Change ((Pre-operative score -post- operative score)±SEM)	Average Percent Change (%±SEM)
NRS (n=18)	3.4±0.6	−45.6±7.7
MPQ (n=15)	3.3±2.0	−6.6±26.4
PCS (n=14)	9.6±2.6	−41.8±10.5
ODI (n=15)	15.7±4.5	−30.2±8.3
BDI (n=15)	5.1±2.2	−15.4±17.4

Next, we examined how patient IONM data correlated with patient outcomes and MRI measurements. Patients who reached MCID on NRS had significantly greater coverage of painful regions with the implanted paddle (r=−0.50, p=0.03). Patients who reached MCID ODI had significantly more coverage of painful regions with both HR-SCS (r=−0.55, p=0.03) and commercial paddles (r=−0.64, p=0.01). Further, in commercial paddles, less stimulation was needed to target these motor responses corresponding with patient dermatome (r=−0.50, p=0.03). Activation of distal muscles with HR-SCS negatively correlated with percentage change in NRS (r= −0.49, p=0.04) and ODI (r= −0.58, p=0.02). The correlations of specific muscles with outcomes and patient specific data are shown in Table 3. APD (r=0.56, p=0.02) as well as interpedicular distance (IPD) (r=0.64, p=0.006) and dorsal CSF thickness (r=0.66, p=0.004) correlated with the max stimulation used. Dorsal CSF (dCSF) thickness also correlates with minimal stimulation amplitude required to obtain a response (r=0.69, p=0.002) and maximum stimulation amplitude (r=0.66, p=0.004).

Table 3:

Correlation coefficients and p-values of maxRMS in muscles related to patient specific data

Paddle	Muscle	Patient specific data	Correlation coefficient (r)	p-value
HR- SCS	AH	50% NRS pain response	−0.49	0.04
	AH	% change in NRS	−0.50	0.04
	AH	% change in ODI	−0.60	0.02
	BF	% change in ODI	−0.63	0.01
	TA	APD	−0.63	0.007
Commercial	AH	% change in NRS	−0.48	0.04
	TA	APD	−0.57	0.017

Discussion

In our study, consisting of 18 patients undergoing SCS surgery with IONM, we looked at the intricate relationships between physiological factors, stimulation parameters, and clinical outcomes of the patients. To address the difference between commercial and HR-SCS paddles, we normalized our heatmaps and scaled with respect to the specific anatomic midline, considering contact and paddle dimensions. For the testing parameters, we only applied IONM stimulation at 60 Hz, which is most commonly used for these types of recording. While these devices are capable of various stimulation waveforms, we did not account for whether patients received burst or high-frequency stimulation in the 3-month postoperative period. Instead, our focus remained on the outcome measures patients reported, regardless of their specific programming parameters. The primary objective of our study was to investigate the correlation between patients who responded positively to the electrophysiological aspects of the stimulation, such as changes in muscle activity in their painful regions, and those who reported overall positive outcomes from SCS treatment.

We found that when evoked motor responses were obtained in muscles corresponding with the patient’s painful dermatomes, there was an improvement at 3 months in both NRS and ODI. It stands to reason that if stimulation is reaching the patient’s painful area, the intensity of pain and the patient’s disability will be reduced²⁶. Frankly, the question is why all outcome measures do not respond in kind. First, it must be considered that the Likert RMS score we provide is a global assessment and while it may cover some painful areas well, it may not cover all of them. Many of our patients reported pain outside the area where SCS would be expected to cover and despite instructions given by a member of our medical team to only consider the pain that is being treated by SCS, this remains difficult.

We next examined whether activation of a specific muscle during IONM led to better outcomes. We observed a negative correlation between max RMS in AH on HR-SCS paddle and NRS pain score. This finding is interesting as it is rare that the pain in the distribution of the AH is the sole reason someone undergoes SCS. However, the AH is innervated by branches of the tibial nerve, which originates from the sciatic nerve. Also innervated by the sciatic nerve or its branches are the BF (biceps femoris), medial gastrocnemius (tibial) and the TA (deep peroneal nerve). BF correlated with improvements in ODI, while no correlations were seen with MG or TA. Though significant, the correlations are somewhat modest, and we need to confirm them in a larger data set. Further, it may be interesting in the future not only to ask patients where their pain is related to their feet, but also to carefully check sensation and ask about numbness, paresthesia, and tingling if AH coverage is indeed shown in the future to be a biomarker in a larger sample size.

In addition to evoked motor responses, we add to the literature on patient specific MR findings and stimulation requirements. Patients with smaller APDs were likely to require less energy and have better outcomes on all 3 metrics at 3 months. Notably, APD considers dorsal CSF thickness in its measurement. Previously, we have shown that dorsal CSF thickness correlated with energy requirements on post-operative day one programming amplitude using tonic stimulation¹⁹. Others have shown supporting data that CSF thickness is a notable variable in amplitudes needed to evoke responses and potentially to treat patients^{27, 28}. Here we show that dorsal CSF thickness correlated with minimal and maximal amplitude to response and RMS scores in many muscle groups. Lempka’s group found with modeling that the amplitude of evoked compound action potentials (ECAPs) correlated similarly with dorsal CSF thicknesses²⁹. This is particularly interesting as ECAPs are a measure of therapy reaching the dorsal column as its target. Data has suggested that the more time the therapy elicits ECAPs, the higher dose of therapy the patient receives³⁰. As dorsal CSF thickness is a portion of APD, with the bony architecture making up the remainder, it is not surprising that the closer the electrode is to the area it is targeting, the better the response. The correlation between outcomes and APD, but not between outcomes and CSF thickness, is not unexpected as dorsal CSF thickness changes with posture but bony distances do not³¹. Intraoperatively, CSF thickness remains stable as the patient remains in the prone position and the lead is not subject to forces of patient movement.

Our study has several limitations. Most notably, this was an NIH-sponsored trial, and patient recruitment was limited to 18 patients. The number and heterogeneity of patients limit the statistics we can perform, especially as we assign patients to different subgroups. We show that motor evoked responses in painful regions correlated with outcomes. We acknowledge that APD may be a notable confounder of this data and in a larger subset of data it would be worth performing multi-variate analysis. Our previous work demonstrated that HR-SCS induced distinct patterns in lower extremities with higher precision^{13, 32}. While we noted a clear medio-lateral selectivity over the dorsal column, which was mapped based on these evoked muscle responses, we did not find a significant correlation between these clinical features and muscle responses. One reason might be the small number of subjects compared to high dimensional EMG dataset and high variance in pain location. Thus, these data should be considered preliminary findings for further exploration. Interestingly, patients who had pain for a longer duration showed more improvement in NRS in contrast to the literature. SCS has been shown to have a success rate of 85% if implantation takes place within two years of pain symptoms appearing³³. It is important to note that this study only looked at short-term effects of SCS implantation and NRS scores were assessed 3 months after surgery.

Conclusion

For the first time, we discuss how coverage of a patient’s painful dermatomes via evoked responses intraoperatively correlates with outcomes at 3 months.