Treatment effect modifiers for individuals with acute low back pain: secondary analysis of the TARGET trial

Abstract

Treatment effect modifiers identify patient characteristics associated with treatment responses. The purpose of this secondary analysis was to identify potential treatment effect modifiers for disability from the TARGET trial that compared usual care (control) with usual care + psychologically informed physical therapy (PIPT). The sample consisted of a STarT Back tool identified high-risk patients with acute low back pain that completed Oswestry Disability Index (ODI) data at index visit and 6 months later (n = 1250). Candidate treatment effect modifiers were identified a priori and informed by the literature. Linear mixed models tested for treatment effect modification through tests of statistical interaction. All statistical interactions (P ≤ 0.20) were stratified by modifier to inspect for specific effects (P ≤ 0.05). Smoking was identified as a potential effect modifier (treatment * smoking interaction, P = 0.08). In participants who were smokers, the effect of PIPT was (ODI = 5.5; 95% CI: 0.6-10.4; P = 0.03) compared with usual care. In participants who were nonsmokers, the effect of PIPT was (ODI = 1.5; 95% CI: −1.4 to 4.4; P = 0.31) compared with usual care. Pain medication was also identified as a potential effect modifier (treatment × pain medication interaction, P = 0.10). In participants prescribed ≥3 pain medications, the effect of PIPT was (ODI = 7.1; 95% CI: −0.1 to 14.2; P = 0.05) compared with usual care. The PIPT effect for participants prescribed no pain medication was (ODI = 3.5; 95% CI: −0.4 to 7.4; P = 0.08) and for participants prescribed 1 to 2 pain medications was (ODI = 0.6; 95% CI: −2.5 to 3.7; P = 0.70) when compared with usual care. These findings may be used for generating hypotheses and planning future clinical trials investigating the effectiveness of tailored application of PIPT.

1. Introduction

Low back pain (LBP) is a leading cause of disability worldwide and foremost condition for rehabilitation services.⁹ In contrast to many randomized clinical trials investigating treatment efficacy for LBP, there are fewer investigations of treatment effect moderation. Treatment effect modifier studies are important because of extensive interindividual variability of LBP,¹⁵ with many factors potentially affecting treatment response. Treatment effect modifiers identify patient characteristics associated with treatment responses. They can be used to guide treatment delivery or generate hypotheses for future research. Detecting such patient characteristics is a top research priority for LBP.^10,13 This effort is part of growing focus on precision medicine—identifying characteristics that predict treatment response to provide the right care to the right patient at the right time.²⁴

Previous studies have identified several potential treatment effect modifiers for common LBP treatments. A recent meta-analysis pooled data from 27 randomized trials of persistent nonspecific LBP (n = 3514) and identified not having physical work demands, medication use, and lower body mass index as potential treatment effect modifiers for exercise.²⁷ A separate systematic review generated from 4 randomized trials of nonspecific LBP (n = 7208) identified age, employment factors, back pain status, opioid medication, treatment expectations, and education as potential treatment effect moderators for physical therapist–delivered interventions.²³ Secondary analysis of the STarT Back Trial results identified socioeconomic status, education, and number of pain medications as potential treatment effect modifiers of stratified care for LBP.²

In this study, we investigated potential treatment effect modifiers in a novel patient population (ie, those with acute LBP at high risk for persistent disability). Ideally, treatment effect moderation is tested within a parallel arm, randomized clinical trial that can distinguish between general prognostic factors (predictors of outcome regardless of treatment) and factors predictive of positive response to specific treatment (treatment effect moderators).^25,26,34,50 The clinical trial should consist of a large enough sample size to support secondary analyses of treatment effect moderation, despite not being specifically powered to detect treatment effect moderators.^5,34 The TARGET trial provides several methodological strengths for investigating treatment effect modifiers. It was a large, pragmatic, cluster randomized trial that tested the effectiveness of early referral to psychologically informed physical therapy (PIPT) in preventing STarT Back identified high-risk patients with acute LBP from transitioning to chronic LBP.^11,12 The TARGET trial was pragmatic and thus associated with a high degree of generalizability (ie, external validity) to the general population of persons with acute LBP presenting to primary care.³³ Referrals for PIPT occurred through an automatic best practice alert in the electronic medical record triggered by high-risk scores from the STarT Back tool.^11,12 The comparator group received usual care. TARGET trial primary results analysis found no group differences in rates of transition to chronic LBP and patient-reported disability at 6 months.¹² Patient-level characteristics that could serve as treatment effect modifiers have not yet been investigated in the TARGET trial. Identification of such characteristics could drive hypothesis generation for future research efforts, including clinical trial planning. The purpose of this secondary analysis was to identify potential treatment effect modifiers from the TARGET trial.

2. Methods

2.1. Study design and setting

The study protocol of TARGET (Targeted Interventions to Prevent Chronic Low Back Pain in High-Risk Patients; ClinicalTrials.gov, NCT02647658) has been published.¹¹ In brief, clusters were primary care clinics that were randomized to usual care + PIPT (intervention) or usual care (control) groups. Patients presenting to primary care clinics were identified as currently having acute LBP based on the NIH Task Force 2-item Chronic LBP Questionnaire.¹³ Patients were considered as having acute LBP if they reported that their LBP was less than 3 months in duration or if they experienced LBP for less than half of the days for duration longer than 3 months. Patients with acute LBP who were identified as high risk for persistent symptoms using the STarT Back tool (psychosocial subscore ≥ 4)²⁹ were eligible for TARGET trial enrollment. Trial participants were assessed for chronic LBP status and self-reported disability 6 months later as dual primary outcomes. Patients were enrolled between May 2016 and June 2018 in 77 primary care practices in 4 U.S. health systems (Pittsburgh, PA, Boston, MA, Salt Lake City, UT, and Baltimore, MD), and the follow-up was completed by March 2019.

2.2. Participants

This planned secondary analysis includes patients enrolled in the TARGET cluster randomized clinical trial and completed patient-reported disability data at index visit and 6 months later.

2.3. Procedures

TARGET trial procedures have been reported elsewhere,^11,12 with interventions briefly described below.

2.3.1. Usual care (control)

For all participating primary care clinics, before recruitment, we used educational outreach through grand rounds, staff meetings, and online modules to review the acute LBP guidelines without monitoring or incentivizing adherence. Education was based on the most up-to-date guidelines⁸ and included recommendations for a focused history and physical examination, with diagnostic imaging and testing not indicated unless signs of severe or progressive neurologic deficits or serious nonmusculoskeletal conditions were present or suspected.¹¹

2.3.2. Usual care + psychologically informed physical therapy (intervention)

Clinics randomized to the intervention received enhanced usual care with specific instructions for immediate referral of high-risk patients to a stratified approach to care.^16,29,30 Psychologically informed physical therapy training programs were delivered to physical therapists in each geographical region who commonly receive referrals from the primary care clinics randomized to the intervention.¹ In brief, course content provided an overview of theoretical rationale and supporting data for PIPT, specific management principles and skills, with demonstration and practice, and treatment monitoring components. Psychologically informed physical therapy focused on educating patients about their LBP condition, reducing fear of movement, and improving coping skills, as well as addressing physical impairments, such as spinal mobility deficits and pain. Owing to the pragmatic nature of the trial, standardization of PIPT delivery was not closely monitored. However, we developed a PIPT checklist to be included with the physical therapist referral that included key components of PIPT.^1,11

2.4. Description of candidate treatment effect modifiers

Credibility of our subgroup analysis was strengthened by incorporating criteria from the approach recommended by Sun et al⁵⁰ which included investigating a limited number of prespecified effect modifier variables, prespecification of the hypothesized direction of subgroup effects, and use of interaction terms. Several patient variables (ie, socioeconomic status, smoking, and obesity) were a priori identified as per the clinical trial registry (ClinicalTrials.gov Identifier: NCT02647658) to explore heterogeneity of treatment effects by examining statistical interactions between treatment and potential modifiers.^18,44,46 Selection of additional variables as candidate treatment effect modifiers for this secondary analysis was based on updated LBP treatment moderator literature, prognostic cohort studies, or theoretical support. Candidate treatment effect modifiers are presented in Table 1.

Table 1.

Candidate treatment effect modifiers.

Variable	Description	Literature support	Hypothesized direction of larger effect from PIPT*
Age (y)	≤44	Gurung, et al. (2015)23	Younger > older
	45-64	Garcia, et al. (2015)19
	≥65
Sex	Male	Gurung, et al. (2015)23	Female patients > male patients
	Female	Fillingim (2017)15
Race	White	Milani, et al. (2018)39	White > Black, Other
	Black	Fillingim (2017)15
	Other
Ethnicity	Hispanic	Milani, et al. (2018)39	Non-Hispanic > Hispanic
	Non-Hispanic	Fillingim (2017)15
		Campbell & Edwards (2012)6
Insurance†	Medicare	Katzan, et al. (2019)32	all > Medicaid
	Medicaid
	Private
	Self-pay
	Other
SES‡†	ADI quartiles	Beneciuk, et al. (2017)2	Low > high SES
	1, low deprivation	Campbell, et al. (2013)7
	2	Moffett, et al. (2009)40
	3
	4, high deprivation
BMI†	≥30 (obese)	Hayden, et al. (2020)27	Low > high BMI
	<30 (not obese)
Smoking†	Smoker	de Zoete, et al. (2020)57	Nonsmoker > smoker
	Nonsmoker	Roseen, et al. (2021)47
LBP location§	Axial LBP	Traeger, et al. (2016)51	axial LBP > LBP and leg pain
	LBP and leg pain	Petersen, et al. (2015)43
		Hider, et al. (2015)28
		Konstantinou, et al. (2013)36
Pain medication	0	Gurung, et al. (2015)23	Less > more pain medication
	1-2	Beneciuk, et al. (2017)2
	≥3	Hayden, et al. (2020)27
		Roseen, et al. (2021)47

^*Hypothesized direction of the interaction effect (ie, usual care + PIPT associated with larger effects compared with usual care alone for patients).

^†Candidate treatment effect modifiers identified a priori.

^‡Socioeconomic status was categorized into quartiles based on the geographic location of clinic sites and the national Area Deprivation Index (ADI).[35,53] The ADI ranks census block or neighborhood for socioeconomic disadvantage, with higher ADI scores indicating greater disadvantage. The validated Neighborhood Atlas tool was used to estimate national-level ADI scores for each clinic location and then divided the distribution into quartiles.[49]

^§LBP location based on ICD-9 or ICD-10-CM diagnostic codes present at the index encounter.

ADI, Area Deprivation Index; BMI, body mass index; LBP, low back pain; SES, socioeconomic status; PIPT, psychologically informed physical therapy.

2.5. Outcome measure

The TARGET trial primary outcome measures were the proportion of patients who transitioned to chronic LBP¹³ and patient-reported disability at 6 months. No between-group differences in primary outcomes were detected in the primary analysis.¹² In this analysis to identify potential treatment effect modifiers for PIPT, we only considered patient-reported disability as an outcome because it was measured on a continuous scale (ie, chronic LBP through NIH definition is a binary outcome) and would be more sensitive to 6-month changes.

Low back pain disability was assessed during index visit and 6 months later with the Oswestry Disability Index (ODI), which has 10 items that assess how LBP affects common daily activities.¹⁷ The ODI has a range of 0% “no disability due to LBP” to 100% “completely disabled due to LBP,” with higher scores indicating higher LBP-related disability. The ODI has been found to have sound psychometric properties with a minimal clinically important difference reported to be 10 percentage points.^17,41 Previous recommendations and clinical trials have suggested 50% improvement in patient-reported outcomes as a clinically relevant threshold.^14,41

2.6. Data analysis

Baseline demographic and clinical characteristics were summarized with frequencies and percentages for categorical variables and means and standard deviations for the continuous variable ODI at baseline. Our secondary analysis included only those individuals with baseline and 6-month follow-up; therefore, we assessed whether the balance on baseline variables from randomization was relatively preserved. We tested for differences in categorical baseline variables with chi-squared tests adjusted for clustering by clinic with Taylor series linearization for variance estimation and for baseline ODI with a linear mixed model with a fixed site effect and random clinic effect. We used linear mixed models with fixed effects for baseline ODI, site, the potential treatment effect modifier of interest, treatment group, treatment × modifier interaction, and a random clinic effect to control for the cluster randomized design of the study to test for treatment effect modification on the ODI score at 6 months. For treatment × modifier interactions with P value ≤ 0.20, we estimated treatment effects, corresponding 95% confidence intervals, and Cohen d stratified by the modifier, similar to previous studies investigating treatment effect modification.^2,23,47 Stratifying by the modifier allowed for inspection of specific effects with a P value of ≤ 0.05 used as the criterion for determining differences between treatment conditions.

An additional exploratory analysis was conducted using a dichotomized effect of 50% improvement in ODI scores at 6 months. This effect was tested using a generalized linear mixed model with a logit link and the same fixed effects and random effects from the model for a continuous 6-month ODI score.

3. Results

3.1. Sample characteristics

Baseline characteristics of high-risk patients enrolled in the TARGET trial and providing ODI data at index visit and 6 months later (n = 1250) are presented in Table 2. These individuals were predominantly White (78%), non-Hispanic (92%), having private insurance (43%), diagnosed with axial LBP (68%), and 1 to 2 pain medication prescriptions at index visit (62%). Treatment groups were balanced on all baseline measures. Baseline ODI scores were similar for high-risk patients allocated to intervention (48.0 ± 16.9) or control (47.8 ± 17.5) groups (P = 0.80).

Table 2.

Baseline characteristics of sample^*.

Variable	Total sample (n = 1250)	Usual care (n = 622)	Usual care + PIPT (n = 628)	P
Age (≤44)	539 (43)	265 (43)	274 (44)	0.16
45-64	476 (38)	222 (36)	254 (40)
65+	235 (19)	135 (22)	100 (16)
Sex (female)	767 (61)	384 (62)	383 (61)	0.80
Race (White)	969 (78)	460 (74)	509 (81)	0.40
Black	206 (16)	113 (18)	93 (15)
Other	37 (3)	26 (4)	11 (2)
Missing	38 (3)	23 (4)	15 (2)
Ethnicity (non-Hispanic)	1144 (92)	573 (92)	571 (91)	0.46
Hispanic	70 (6)	37 (6)	33 (5)
Missing	36 (3)	12 (2)	24 (4)
Insurance (Medicare)	218 (17)	113 (18)	105 (17)	0.92
Medicaid	168 (13)	77 (12)	91 (14)
Private	541 (43)	257 (41)	284 (45)
Self-pay	113 (9)	56 (9)	57 (9)
Other	47 (4)	27 (4)	20 (3)
Missing	163 (13)	92 (15)	71 (11)
ADI (Q1, low deprivation)	359 (29)	148 (24)	211 (34)	0.83
Q2	195 (16)	105 (17)	90 (14)
Q3	320 (26)	186 (30)	134 (21)
Q4, high deprivation	376 (30)	183 (29)	193 (31)
Obesity (yes)	617 (49)	300 (48)	317 (50)	0.25
No	589 (47)	294 (47)	295 (47)
Missing	44 (4)	28 (5)	16 (3)
Current smoker (yes)	225 (18)	103 (17)	122 (19)	0.81
No	833 (67)	423 (68)	410 (65)
Missing	192 (15)	96 (15)	96 (15)
LBP location (axial)	855 (68)	418 (67)	437 (70)	0.63
LBP and leg pain	395 (32)	204 (33)	191 (30)
Pain medications (0)	383 (31)	197 (32)	186 (30)	0.78
1 or 2	772 (62)	377 (61)	395 (63)
3+	95 (8)	48 (8)	47 (7)
ODI (baseline)	47.9 (±17.2)	47.8 (±17.5)	48.0 (±16.9)	0.80

Data presented as frequency counts (percentage) unless otherwise indicated.

^*SBT high-risk participants with ODI data at baseline and 6 months.

ADI, Area Deprivation Index; LBP, low back pain; ODI, Oswestry Disability Index; PIPT, psychologically informed physical therapy.

3.2. Potential treatment effect modifiers

No treatment effect moderation was detected for age, sex, race, ethnicity, insurance, and LBP location. SES, BMI, smoking status, and pain medication prescription at intake visit were identified as potential treatment effect modifiers for disability scores at 6 months through tests of statistical interaction (P’s ≤ 0.20) (Table 3). These statistical interactions were stratified by the modifier using previously described criterion (P ≤ 0.05) to investigate specific effects within different levels of each potential treatment effect modifier. Stratification by SES (treatment × SES interaction, P = 0.19) or BMI (treatment × BMI interaction, P = 0.16) revealed no specific effects within levels of the modifying factor (P’s > 0.05).

Table 3.

Oswestry Disability Index score at 6 months adjusted for baseline score and study design.

	Usual care			Usual care + PIPT			Stratified by modifier			Model	Interaction test
	n	${\overline{\mathit{x}}}_1$	95% CI	n	${\overline{\mathit{x}}}_2$	95% CI	${\overline{\Delta }}_{1-2}$	95% CI	P	R2*	Partial R2† (95% CI)	P ‡
Age (<=44)	265	29.0	(26.4, 31.7)	274	26.5	(24.0, 29.1)	2.5	(−1.1, 6.1)	0.17	0.18	0.0021 (0.0000, 0.0077)	0.52
45-64	222	30.7	(27.9, 33.5)	254	28.9	(26.2, 31.5)	1.8	(−1.9, 5.5)	0.33
65+	135	31.8	(28.4, 35.2)	100	30.6	(26.8, 34.5)	1.1	(−3.8, 6.1)	0.65
Sex (Male)	238	27.2	(24.5, 30.0)	245	26.0	(23.3, 28.7)	1.2	(−2.5, 4.9)	0.51	0.19	0.0022 (0.0000, 0.0092)	0.24
Female	384	31.8	(29.5, 34.1)	383	29.1	(26.8, 31.4)	2.7	(−0.5, 5.8)	0.09
Race (White)	460	28.7	(26.1, 31.3)	509	26.8	24.5, 29.2)	1.9	(−1.3, 5.0)	0.24	0.16	0.0012 (0.0000, 0.0055)	0.61
Black	113	32.5	(28.8, 36.3)	93	30.3	(26.2, 34.3)	2.2	(−3.2, 7.7)	0.42
Other	26	28.6	(21.8, 35.4)	11	27.4	(17.2, 37.7)	1.2	(−11.1, 13.5)	0.85
Ethnicity (non-Hispanic)	573	29.6	(27.5, 31.8)	571	28.1	(26.0, 30.3)	1.5	(−1.3, 4.3)	0.30	0.18	0.0017 (0.0000, 0.0082)	0.39
Hispanic	37	34.8	(29.0, 40.6)	33	30.4	(24.3, 36.5)	4.3	(−4.0, 12.7)	0.30
Insurance (Medicaid)	77	37.3	(33.2, 41.5)	91	35.3	(31.4, 39.2)	2.0	(−3.5, 7.5)	0.47	0.25	0.0041 (0.0000, 0.0099)	0.49
Medicare	113	35.0	(31.5, 38.5)	105	33.1	(29.4, 36.8)	1.9	(−3.0, 6.7)	0.45
Other	27	36.7	(30.1, 43.2)	20	28.2	(20.7, 35.7)	8.5	(−1.4, 18.4)	0.09
1Private	257	25.5	(22.8, 28.1)	284	24.1	(21.6, 26.6)	1.3	(−2.1, 4.8)	0.44
Self-pay	56	34.6	(29.3, 39.9)	57	31.4	(25.7, 37.1)	3.2	(−4.1, 10.5)	0.39
ADI (Q1, low deprivation)	148	29.6	(25.8, 33.4)	211	25.7	(22.3, 29.1)	3.9	(−1.1, 8.9)	0.12	0.19	0.0058 (0.0000, 0.0140)	0.19
Q2	105	27.0	(22.6, 31.5)	90	30.5	(25.4, 35.5)	−3.4	(−10.1, 3.2)	0.31
Q3	186	33.9	(29.7, 38.1)	134	29.3	(24.9, 33.7)	4.6	(−1.2, 10.4)	0.11
Q4 (high deprivation)	183	30.5	(26.0, 35.0)	93	29.6	(25.8, 33.4)	0.9	(−4.5, 6.3)	0.75
BMI < 30	294	28.3	(25.7, 30.9)	295	27.5	(24.9, 30.0)	0.8	(−2.6, 4.3)	0.62	0.19	0.0030 (0.0000, 0.0111)	0.16
≥30	300	31.9	(29.4, 34.5)	317	28.7	(26.2, 31.1)	3.3	(−0.1, 6.6)	0.06
Current smoker (no)	423	28.9	(26.5, 31.2)	410	27.4	(25.1, 29.7)	1.5	(−1.4, 4.4)	0.31	0.20	0.0044 (0.0000, 0.0145)	0.08
Yes	103	37.3	(33.4, 41.0)	122	31.7	(28.1, 35.3)	5.5	(0.6, 10.4)	0.03
Diagnosis (axial LBP)	418	29.6	(27.3, 31.9)	437	27.0	(24.7, 29.2)	2.6	(−0.4, 5.7)	0.09	0.18	0.0030 (0.0000, 0.0110)	0.24
LBP and leg pain	204	31.3	(28.3, 34.2)	191	30.6	(27.6, 33.5)	0.7	(−3.3, 4.6)	0.73
Pain medication (0)	197	34.0	(31.1, 36.8)	186	30.5	(27.6, 33.3)	3.5	(−0.4, 7.4)	0.08	0.20	0.0048 (0.0000, 0.0133)	0.10
1 or 2	377	28.2	(25.9, 30.6)	395	27.6	(25.3, 29.9)	0.6	(−2.5, 3.7)	0.70
3+	48	27.9	(22.7, 33.1)	47	20.9	(15.7, 26.0)	7.1	(−0.1, 14.2)	0.05

Δ = ODI mean difference = usual care—(usual care + PT).

^*R² is for the complete linear mixed model.

^†Partial R² for the interaction term only.

^‡P value for partial R2.

ADI, Area Deprivation Index; BMI, body mass index; LBP, low back pain; PIPT, psychologically informed physical therapy.

Smoking was identified as a potential effect modifier (treatment × smoking interaction, P = 0.08). In participants who were smokers, the effect of PIPT was (ODI = 5.5; 95% CI: 0.6-10.4; P = 0.03; d = 0.28) compared with usual care. In participants who were nonsmokers, the effect of PIPT was (ODI = 1.5; 95% CI: −1.4-4.4; P = 0.31; d = 0.07) compared with usual care (Table 3). Unadjusted ODI 6-month change scores by smoking status across treatment groups are presented in Figure 1.

Figure 1.

Unadjusted Oswestry Disability Index 6-month change scores by smoking status across treatment groups. Box plot length represents interquartile range (distance between 25th and 75th percentiles). Numerical value represents group mean. Horizontal line represents group median. Whiskers represent group minimum and maximum values. PIPT, psychologically informed physical therapy; UC, usual care.

Pain medication was also identified as a potential effect modifier (treatment * pain medication interaction, P = 0.10). In participants prescribed ≥3 pain medications, the effect of PIPT was (ODI = 7.1; 95% CI: −0.1 to 14.2; P = 0.05; d = 0.34) compared with usual care. The effect of PIPT for participants prescribed no pain medication was (ODI = 3.5; 95% CI: −0.4 to 7.4; P = 0.08, d = 0.19) and for participants prescribed 1 to 2 pain medications was (ODI = 0.6; 95% CI: −2.5 to 3.7; P = 0.70, d = 0.03) when compared with usual care (Table 3). Unadjusted ODI 6-month change scores by pain medication prescription across treatment groups are presented in Figure 2.

Figure 2.

Unadjusted Oswestry Disability Index 6-month change scores by pain medication prescription across treatment groups. Box plot length represents interquartile range (distance between 25th and 75th percentiles). Numerical value represents group mean. Horizontal line represents group median. Whiskers represent group minimum and maximum values. PIPT, psychologically informed physical therapy.

3.3. Additional exploratory analysis

Sex, ethnicity, and SES were identified as potential treatment effect modifiers for achieving 50% improvement in disability scores at 6 months with statistical evidence (P ≤ 0.20). Stratification by ethnicity (treatment × ethnicity interaction, P = 0.14) indicated an effect among patients of Hispanic ethnicity (adjusted OR = 3.09, 95% CI = 1.00-9.50, P = 0.05). Among Hispanics, those in the PIPT group were more likely to achieve 50% ODI improvement at 6 months compared with those in the usual care group (45% vs 22%). Among non-Hispanics, those in either the PIPT or usual care group were equally as likely to achieve 50% ODI improvement at 6 months (44% vs 44%). Stratification by sex (treatment * sex interaction, P = 0.17) and by SES (treatment * SES interaction, P = 0.05) revealed no specific effects (P’s > 0.05).

4. Discussion

The purpose of this secondary analysis was to identify potential treatment effect modifiers from the TARGET trial for patients at high risk for poor disability outcomes at 6 months. We identified several patient characteristics that could potentially serve as treatment effect modifiers for PIPT including SES, BMI, smoking status, and pain medication prescription. Follow-up analyses indicated smoking and pain medication were associated with the largest observed treatment effects for STarT Back determined high-risk patients with acute LBP receiving PIPT, compared with usual primary care. Current smokers or individuals prescribed ≥3 pain medications in PIPT were associated with less disability at 6 months compared with those in the usual care; however, these observed effects were modest and their direction opposite to our hypotheses (Table 1), potentially weakening credibility of a real subgroup effect. These hypotheses were based on studies that did not include patients with acute LBP and were at high risk for future disability and that could be another reason for the difference in direction. Regardless, differences between observed and hypothesized effects for treatment modification are not unexpected and reflect challenges of empirical investigation of personalized pain management.²² For example, converging with these findings from the TARGET trial, the SOPPRANO trial, reported opposite effects to what had been hypothesized for response to propranolol for temporomandibular pain after planned subgroup analysis.⁴⁸ LBP studies have reported similar results with treatment effects observed in opposite direction to authors’ hypothesis.¹⁹ Moreover, moderating effects observed in this current analysis were not large enough to provide recommendations for PIPT in clinical practice, although do provide data for hypothesis generation and future clinical trial planning. Collectively, direct comparisons with existing literature are difficult because our sample was isolated to patients with acute LBP at high risk for poor disability outcomes and receiving PIPT.

4.1. Effect modification from pain medication and smoking status

Previous literature involving pain medication as a treatment effect modifier is inconsistent.^2,23,27,47 Our findings provide preliminary evidence that high-risk patients may benefit from multimodal treatment approaches (eg, PIPT + pain medication if appropriate); however, the LBP guidelines do not provide clear direction for how to combine multimodal treatments.⁴⁵ Best practice recommendations emphasize individualized, multimodal, and multidisciplinary aspects of patient-centered care,⁵⁴ which may be most relevant for individuals at high risk for persistent pain.⁵⁶ Nonsmoking status has been associated with positive treatment moderating effects for spinal manipulation,⁵⁷ which did not coincide with our findings related to PIPT. High-risk patients who were current smokers and in PIPT were associated with larger treatment effects at 6 months compared with those in the usual care. This finding potentially suggests that there may have been benefit for smokers in PIPT; however, these observed effects were modest, and nonsmokers were associated with larger treatment effects overall.

Collectively, our findings are preliminary and contrast other observations regarding smoking^47,57 and pain medication^2,23 as potential LBP treatment effect modifiers. Future research needs to consider whether these factors are indirect measures for other factors that may affect response to PIPT. For example, converging evidence indicates pain and nicotine or tobacco use interact through complex mechanisms to influence outcomes and response to treatment with negative affect, coping, and self-efficacy emerging as potential contributors.³⁷

4.2. Effect modification from socioeconomic status

We also detected modest effects between treatment and SES in both primary and exploratory analyses; however, stratification by the SES level was inconsistent and difficult to interpret which may increase likelihood of a spurious finding. Previous studies investigating SES influence on LBP outcomes have not focused on high-risk patients, which make comparisons with prior literature difficult.³² We used clinic site location and the Area Deprivation Index (ADI)^35,53 as an indirect measure of SES, which accurately classifies a community, but may not reflect socioeconomic disadvantage on an individual level. SES is a multidimensional construct that includes education, occupation, and income; therefore, determining which specific dimension(s) are associated with differential treatment effects will be an important aspect to matching patients with most appropriate treatment. Collectively, based on our findings and those from previous studies,^2,7,40,55 future investigation of SES at a granular level as a potential treatment effect modifier for LBP is suggested.

4.3. Comparison with previous low back pain treatment effect modification literature

The TARGET trial provided several methodological strengths for investigating treatment effect modification including a parallel arm, randomized clinical trial study design and large sample size.^34,50 Our findings provide an important contribution to literature because this is the first treatment effect modification analysis involving a highly generalizable sample of patients with acute LBP at high risk for persistent disability and receiving PIPT. Previous LBP treatment effect modification analyses did not exclusively recruit individuals with acute LBP, relying instead on samples including individuals with chronic LBP. Previous analyses also did not exclusively include patients classified as STarT Back high risk, instead identified treatment effect modification for risk stratified care in general (ie, low, medium, and high risk together).²

Focusing on high-risk patients may provide better guidance on treatment matching for PIPT in comparison with previous studies that have incorporated a mix of STarT Back risk groups. PIPT is perhaps “most” aligned with STarT Back high-risk patients because these individuals tend to be associated with elevated yellow flags³ and complex clinical profiles with poor outcome trajectories when compared with those identified as low or medium risk.^20,29,49 Moreover, previous treatment effect moderation studies have primarily focused on “traditional” interventions for LBP^27,57 with fewer studies investigating factors that influence PIPT response (a more complex treatment approach). For example, a previous study investigating physical therapy and graded exercise or exposure for acute LBP reported larger treatment effects for those that had no strong treatment preferences; however, the results were not generated through formal treatment effect modification analysis.²¹

4.4. Study limitations

There are several limitations to this study. A major limitation of the TARGET trial was a 40% nonresponse rate for primary outcomes at 6 months; however, these rates did not differ between the intervention and usual care group, and models using imputation for the missing values did not alter the primary findings.¹² Standards for nonresponse rates have not yet been established for pragmatic trials; therefore, determining impact on generalizability is difficult because best practice recommendations relative to these “real-world” study designs continue to emerge.³⁸ Second, we did not include other potentially relevant treatment effect modifiers in this analysis. For example, other socioeconomic indicators (eg, education, household income, and job occupation) have been included in similar analyses and identified as potential stratified care treatment effect moderators²; however, this information was not collected in the TARGET trial. Moreover, patient treatment expectations have been identified as potential treatment effect moderators in other clinical trials for LBP and chronic pain.^4,47,52 Treatment expectations were not assessed in the TARGET trial; therefore, they were not able to be included in this secondary analysis. Third, although the STarT Back tool was used to identify high-risk patients (primarily based on responses to psychological items), specific unidimensional psychological measures (eg, anxiety, depression, and pain catastrophizing) were not collected which may have further informed our findings particularly related to PIPT. Fourth, PIPT treatment was not standardized and delivered as a component to a pragmatic trial; therefore, definitive characterization of treatment is not possible. There is a need to develop and efficiently assess PIPT treatment fidelity during routine clinical practice, so this treatment approach can be thoroughly investigated in future pragmatic trials. Fifth, treatment modification effects observed must be interpreted as preliminary. These findings are useful to guide future research; however, they are not sufficient for guiding clinical practice in pursuit of treatment–patient matching. Finally, although we prespecified variables that would be investigated in secondary analysis as treatment effect modifiers in the registered statistical analysis plan, the investigation was not based on a strong theoretical model which may increase the likelihood of spurious findings.

4.5. Future research opportunities

The results from this secondary analysis provide several opportunities worthy of future investigation. Although investigating factors predictive of PIPT response are important to inform future analyses, studies incorporating formal treatment moderator analysis are needed if one of the overarching goals is to improve PIPT delivery during clinical practice and as a component to future pragmatic clinical trials. Our findings related to smoking and pain medication prescription as treatment effect modifiers were unexpected and need to be confirmed in similar LBP samples to determine whether these are indirect measures for other factors that may affect response to PIPT or actual treatment effect modifiers. For example, those who smoke may receive more benefit from PIPT primarily because of behavior-based components as many patients experiencing pain identify smoking as a pain coping mechanism.^31,42 Pain medication as a potential treatment effect modifier is particularly intriguing because this has been a consistent finding in previous LBP literature^2,23,27,47; however, direction of effect has been inconsistent, and most clinical trials (including the TARGET trial) do not explicitly investigate multimodal treatment approaches. Future clinical trials incorporating more rigorous study design are necessary to determine whether certain patient characteristics (eg, high risk for persistent pain) are associated with treatment effects when receiving pain medication in addition to physical therapy for LBP. Others have suggested that in patients with LBP use of pain medication may increase tolerance for active treatments such as physical therapy in situations where self-management alone may not be sufficient.⁴⁷ Data from large multimodal studies could inform guidelines and provide more clear direction for combined approaches.

5. Conclusion

For this secondary analysis of the TARGET trial, we found no strong evidence of treatment effect modification; however, pain medication prescription and smoking at intake visit were identified as potential treatment effect modifiers. Socioeconomic status and BMI also had some evidence for treatment effect modification, although interpretation of these findings was difficult. Findings from future treatment effect modifier studies may be used to guide tailored application of PIPT; however, these data are best suited for generating hypotheses and planning future clinical trials.