Crosswalking the Patient-Reported Outcomes Measurement Information System Physical Function, Pain Interference, and Pain Intensity Scores to the Roland-Morris Disability Questionnaire and the Oswestry Disability Index

Abstract

Objective:

To link scores from two condition-specific measures for chronic lower back pain (CLBP), the Oswestry Disability Index (ODI) and the Roland-Morris Disability Questionnaire (RMDQ), to Patient Reported Outcomes Measurement Information System (PROMIS^®) physical function, pain interference, and pain intensity scores.

Design:

Ordinary least squares (OLS) regression analyses of existing data to link the PROMIS scores with the ODI and RMDQ.

Setting:

NA

Participants:

Data are from three samples of adults with CLBP: The Center for Excellence in Research for Complementary and Integrative Health (CERC) Study (N=1677); the Assessment of Chiropractic Treatment for Low Back Pain and Smoking Cessation in Military Active Duty Personnel (ACT; N=384); and the pain subsample of the PROMIS 1 Wave 2 Pain and Depression study (PROMIS 1 W2; N=218).

Interventions:

NA

Main Outcome Measures:

PROMIS physical function, pain interference, and pain intensity (CERC, ACT and PROMIS 1 W2); ODI (CERC and PROMIS 1 W2); and RMDQ (ACT and PROMIS 1 W2).

Results:

In predicting PROMIS scores, the ODI model R² values ranged from 0.26-0.56 and the RMDQ model R² values ranged from 0.13-0.50. ODI and RMDQ models were least precise in predicting the PROMIS pain intensity score (R² values ranged from 0.13-0.41) relative to the other PROMIS scores. Models with the three PROMIS scores as predictors yielded R² values ranging from 0.64 – 0.68 and 0.46 – 0.58 for the ODI and RMDQ respectively. Models using combined data from two studies (i.e., PROMIS 1 W2 and ACT, or PROMIS 1 W2 and CERC) tended to be more precise than models using only a single study sample.

Conclusion:

Model results reported here to can be used translate PROMIS physical function, pain interference, and pain intensity scores to and from the ODI and RMDQ. The empirical linkages can facilitate comparisons across CLBP interventions and broaden interpretation of study results.

A variety of patient-reported outcome measures (PROMs) have been used to evaluate the effectiveness of interventions for chronic low back pain (CLBP).^1-10 The Oswestry Disability Index (ODI)¹¹ and the Roland-Morris Disability Questionnaire (RMDQ)¹² are two PROMs used specifically for CLBP. The ODI is focused on disability across a range of domains (e.g., physical function, pain, sleep), while the RMDQ is focused on limitations in physical function due to back pain.¹³

The U.S. National Institutes of Health (NIH) has encouraged routine use of the Patient-Reported Outcomes Measurement Information System (PROMIS^®) measures. PROMIS measures, developed within an item response theory framework,^{14, 15} have demonstrated strong psychometric properties in the general population and across a number of clinical subgroups.¹⁶ The PROMIS-29¹⁷ profile measure assesses pain intensity, physical function, fatigue, pain interference, depressive symptoms, anxiety, ability to participate in social roles and activities, and sleep disturbance.

Although all PROMIS-29 scales are potentially useful, the physical function, pain interference, and pain intensity constructs are most relevant to assess limitations directly due to CLBP and potential improvements in functioning due to therapeutic interventions. PROMIS physical function has also been shown to perform at least as well as the RMDQ among CLBP patients.¹⁸ Additionally, an NIH Research Taskforce proposed the Impact Stratification Score (ISS) as the sum of PROMIS-29 physical function, pain interference, and pain intensity scores.¹⁹

Given the popularity of the ODI and RMDQ,²⁰ the usefulness of PROMIS measures such as the PROMIS-29 will be significantly enhanced with the availability of empirical crosswalks from one set of scores to the other so that CLBP researchers using PROMIS scores can interpret their results in the context of familiar outcomes. Crosswalks will also serve to help CLBP researchers integrate results from studies using the ODI and RMDQ.

As part of a larger study on chronic back pain,²⁰ this paper reports results on a preliminary set of empirical links between the ODI and RMDQ and the three PROMIS scores in the ISS. In addition to facilitating crosswalks for study comparisons, results from this paper can be used in the interpretation of meta-analyses of studies.^{21, 22}

Using data from three large samples of adults with CLBP, we examine associations of the three PROMIS scores (physical function, pain interference, pain intensity) with the ODI and the RMDQ. We then derive equations to predict these PROMIS scores from the ODI and the RMDQ as well as equations to predict the ODI and the RMDQ from the PROMIS scores.

Methods

Data sources

The Center for Excellence in Research for Complementary and Integrative Health (CERC)^{23, 24} data were collected on patients being treated for CLBP and chronic neck pain (CNP; n=2024) and included three subgroups: CLBP only (n=518), CNP only (n=347), and CLBP + CNP (n=1159). For the present study, we use only the baseline data and exclude those indicating the presence of CNP only (n=347) resulting in a sample of 1677. The mean age of the 1677 respondents was 49 (SD=15) ranging from 21 to 95 years of age and 71% female. The sample was predominantly non-Hispanic White (83%) with relatively low rates for Hispanic White (3%), non-Hispanic: Black (2%), Asian (2%), Native Alaskan (2%), and multi-racial (3%); and Other (5%). Assessment of Chiropractic Treatment for Low Back Pain and Smoking Cessation in Military Active Duty Personnel (ACT)^25-27 data (n=750) were collected longitudinally on active military personnel participating in chiropractic clinical trials for low back pain (LBP) and comprised three subgroups: chronic LBP (CLBP; n=384), acute LBP (n=287), and subacute LBP (n=79). This dataset has multiple timepoints, but only the baseline data and the CLBP subgroup are used in this paper. The mean age of the CLBP subgroup of participants at baseline was 32 (SD=8) ranging from 18 to 50 years of age and included 24.7% females. The respondents classified themselves as non-Hispanic: White (53%), Black (18%), Asian (5%), other (3%); Hispanic (18%); or unspecified (2%).

The pain subsample of the PROMIS 1 Wave 2 Pain and Depression study (PROMIS 1 W2):^{16, 28} (n=218) was developed to test the utility of computer adaptive tests (CATs) with individuals diagnosed with back pain with or without sciatica of at least 6 weeks duration who were scheduled for epidural steroid injections. The median age group was 55-59 years and included 56% females. The sample consisted of non-Hispanic: Whites (84%), Blacks (3%), Asian (1%), Native Hawaiian or Pacific Islander (0.5%), multi-racial (6%); Hispanic: Whites (1%), multi-racial (4%); or unknown (1%).

Measures

PROMIS scales – included in CERC, ACT and PROMIS 1 W2 datasets. The CERC and ACT include the PROMIS-29 whereas the PROMIS 1 W2 sample assesses PROMIS scores via CAT.

Physical Function:

Capability of physical activities including functioning of upper and lower extremities, and central regions (neck, back), and instrumental activities of daily living, such as running errands.²⁹ Scores are reported on a T-score metric based on a general population mean of 50 (SD=10). Higher scores indicate better functioning.

Pain Interference:

Consequences of pain on relevant aspects of one’s life. This includes the extent to which pain hinders engagement with social, cognitive, emotional, physical, and recreational activities.³⁰ Scores are reported on a T-score metric based on a U.S. general population mean of 50 (SD=10). Higher scores indicate more pain interference.

Pain Intensity:

A single item measure reflecting how much a person hurts, on average over the past 7 days on a scale from 0-10. Higher scores indicate higher pain intensity. Although the 0–10 item has been collapsed to five categories (0 = 1; 1–3 = 2; 4–6 = 3; 7–9 = 4; 10 = 5) in some prior PROMIS studies,^{31, 32} we preserved the 11 categories of information. For the PROMIS 1 Wave 2 dataset, average pain was asked for both back and leg pain (i.e., In the past 7 days, when you had BACK pain, how would you rate your average pain; and In the past 7 days, when you had LEG pain, how would you rate your average pain). All respondents had either back pain only or both back and leg pain and answered both questions. For the purposes of these analyses, the maximum score of the two was selected.

ODI – included in CERC and PROMIS 1 W2 datasets.

The ODI is a measure of functional disability consisting of 10 items assessing pain intensity, personal care, lifting, walking, sitting, standing, sleeping, sex life (if applicable), social life, and traveling. Response options range from 0 to 5 with higher scores indicating greater disability.

The scale is scored by summing scores across all items, dividing the total score by the maximum possible (e.g., 50 if all 10 items are answered) and then multiplying by 100. The ODI score can be classified as: <20 = minimal disability, 21-40 = moderate, 41-60 = severe, 61-80 = crippled, and 81-100 = bedbound or exaggerating symptoms.¹¹

RMDQ – included in ACT and PROMIS 1 W2 datasets.

The RMDQ is a 24-item measure assessing physical disability due to lower back pain. The items were chosen from the Sickness Impact Profile (SIP)³³ and reflect physical function content directly relevant to lower back pain. Items are dichotomous and yield a total summary score ranging from 0 (no disability) to 24 (maximum disability).

Analyses

The analyses characterize relationships between the three PROMIS scores (physical function, pain interference, pain intensity) and the ODI and RMDQ scores. Following standard recommendations,^{34, 35} a minimum bivariate correlation between scales of 0.80 is required for latent variable equating approaches such as IRT co-calibration. For correlations lower than 0.80, ordinary least squares (OLS) regression models with linear equating^{36, 37} are recommended. OLS models were evaluated in terms of R², and the normalized mean absolute error (NMAE). The NMAE provides information regarding the average deviation between the observed and predicted scores divided by the standard deviation of the observed score. Lower values of the NMAE indicate better performance.

We first evaluated the ODI and RMDQ in predicting the three PROMIS scores (physical function, pain interference, pain intensity). Next, we evaluated the three PROMIS scores in predicting the ODI and RMDQ. We estimated all models within each sample for all respondents with non-missing scores. We also report results for the combined samples to provide a reference for more heterogeneous patient groups with a wide range of disability. Further, we report the full model equations so that they can be applied by researchers to predict PROMIS scores from ODI and RMDQ and to predict ODI and RMDQ from PROMIS scores.

Results

Mean scores and correlations among the PROMs are shown in Tables 1 and 2, respectively, for the three samples individually and combined. Respondents in all three samples report lower than average physical functioning and higher than average pain interference (relative to the U.S. population mean of 50). Notably, the PROMIS 1 W2 sample displays lower physical functioning and higher pain interference than the other two samples. This trend of more disability among the PROMIS 1 W2 sample is also evident in the pain intensity scores, ODI disability scores, and RMDQ scores. When ACT and CERC samples are combined with PROMIS 1 W2, scores reflect more disability than in CERC or ACT alone.

Table 1.

Mean and standard deviation of PROMIS Physical Function, Pain Interference, and Pain Intensity scores, Oswestry Disability Index (ODI) scores, and Roland-Morris Disability Questionnaire (RMDQ) scores for each study sample and for combined datasets

	PF	PI	Avg Pain	ODI	RMDQ
PROMIS I W2 (n=218)	37.5 (5.7)	64.2 (6.6)	6.5 (1.9)	37.4 (15.7)	13.0 (5.6)*
CERC (n=1677)	45.4 (7.2)	55.1 (7.3)	3.9 (2.0)	20.5 (12.8)	NA
ACT (n=384)	44.1 (6.7)	58.9 (6.3)	5.5 (1.9)	NA	9.0 (5.4)
PROMIS + CERC (n=1,895)	44.4 (7.5)	56.3 (7.8)	4.2 (2.2)	22.4 (14.2)	NA
PROMIS + ACT (n=563)	42.0 (7.1)	60.5 (6.8)	5.2 (2.1)	NA	10.3 (5.7)

NOTE: CERC= Center for Excellence in Research for Complementary and Integrative Health; ACT= Assessment of Chiropractic Treatment for Low Back Pain and Smoking Cessation in Military Active Duty Personnel; PROMIS I W2 = PROMIS 1 Wave 2 Pain and Depression study. NA=Not Administered; PF = PROMIS Physical Function T score; PI = PROMIS Pain Interference T score; Avg Pain = PROMIS Pain Intensity score.

^*Only n=179 of the PROMIS 1 W2 sample completed the RMDQ.

Table 2.

Product-moment correlations and sample sizes among PROMIS Physical Function, Pain Interference, and Average Pain Intensity scores, Oswestry Disability Index (ODI) scores, and Roland-Morris Disability Questionnaire (RMDQ) Scores in each of three samples individually and combined.

	PF	PI	Avg Pain	RMDQ PROMIS I W2
ODI PROMIS I W2	−.69 (n=212)	.73 (n=212)	.51 (n=212)	.68 (n=175)
ODI CERC	−.71 (n=1527)	.71 (n=1527)	.59 (n=1674)	NA
ODI CERC+PROMIS 1 W2	−.74 (n=1739)	.75 (n=1739)	.64 (n=1886)	NA
RMDQ PROMIS I W2	−.61 (n=179)	.61 (n=179)	.37 (n=179)	NA
RMDQ ACT	−.69 (n=384)	.67 (n=384)	.36 (n=384)	NA
RMDQ ACT +PROMIS 1 W2	−.71 (n=563)	.69 (n=563)	.45 (n=563)	NA

NOTE: All correlations p<.001. CERC= Center for Excellence in Research for Complementary and Integrative Health; ACT= Assessment of Chiropractic Treatment for Low Back Pain and Smoking Cessation in Military Active Duty Personnel; PROMIS I W2 = PROMIS 1 Wave 2 Pain and Depression study; NA=Not Applicable; PF = PROMIS Physical Function T score; PI = PROMIS Pain Interference T score; Avg Pain = PROMIS Pain Intensity score. PF, PI, and Avg Pain scores are from the samples noted in the corresponding first column (e.g. the PF score that is correlated with the ODI from PROMIS I W2 comes from the PROMIS I W2 sample).

The product-moment correlations among the PROMs (Table 2) show that the ODI and RMDQ are more strongly associated with PROMIS physical function and pain interference than with pain intensity, and this is especially true for the RMDQ. The association between the ODI and RMDQ is close in magnitude to that between the ODI and PROMIS physical function. Also noteworthy is the consistent increase in inter-relationships for combined samples relative to individual samples. This is due to the increase in the range of values from combining datasets. Because none of the correlations meet the 0.80 criterion for latent variable equating, we use OLS with linear equating to link the scales.

Predicting PROMIS scores from the ODI and RMDQ total scores

ODI Total score models

We estimated models predicting the three PROMIS scores from the ODI total score using data from PROMIS 1 W2 and CERC samples (see Table 3). In these models, R² values ranged from 0.26-0.56 with NMAE ranging from 0.57 to 0.75 standard deviations. Models based on combined data performed better than individual sample models, and models predicting PROMIS physical function and pain interference scales were better than those predicting pain intensity. Specifically, for physical function, overall R² ranged from 0.47 – 0.55 and NMAE ranged from 0.58 to 0.63. Similar patterns were observed for pain interference. The ODI score did not perform well in predicting PROMIS pain intensity, where R² values ranged from 0.26-0.41 and NMAE was relatively high (0.66 – 0.75).

Table 3.

Model equation, R² and NMAE for regressions predicting PROMIS Physical Function, Pain Interference, and Average Pain Intensity scores from Oswestry Disability Index for the PROMIS I W2 and CERC samples, individually and combined.

PROMIS Score	Sample	Equation	R2	NMAE
Physical Function	PROMIS I W2 (n=212)	46.762 – 0.249 x ODI total score	0.47	0.63
	CERC (n=1527)	53.622 – 0.401 x ODI total score	0.51	0.60
	CERC+PROMIS 1 W2 (n=1739)	53.216 – 0.389 x ODI total score	0.55	0.58
Pain Interference	PROMIS I W2 (n=212)	52.957 + 0.301 x ODI total score	0.55	0.57
	CERC (n=1527)	46.865 + 0.403 x ODI total score	0.50	0.62
	CERC+PROMIS 1 W2 (n=1739)	47.019 + 0.408 x ODI total score	0.56	0.57
Pain Intensity	PROMIS I W2 (n=212)	4.133 + 0.062 x ODI total score	0.26	0.75
	CERC (n=1674)	1.932 + 0.095 x ODI total score	0.35	0.70
	CERC+PROMIS 1 W2 (n=1886)	1.959 + 0.099 x ODI total score	0.41	0.66

NMAE = Normalized Mean Absolute Error; CERC= CERC= Center for Excellence in Research for Complementary and Integrative Health; PROMIS I W2 = PROMIS 1 Wave 2 Pain and Depression study. R² and NMAE are adjusted using linear equating to address regression to the mean.

Table 3 provides regression results for each model for use in translating scores from one scale to another, although the models predicting PROMIS pain intensity from ODI are limited by the low level of explained variance.

RMDQ Total score models

Results from models predicting the three PROMIS scores from the RMDQ total score using data from PROMIS 1 W2 and ACT samples individually and combined are shown in Table 4. Across all models, R² values ranged from 0.13-0.50 with NMAE ranging from 0.57 to 0.89 standard deviations. Although the pattern of results is like that shown for the ODI models, the PROMIS scores were less accurately predicted by the RMDQ than the ODI. However, like the ODI results, the RMDQ was a stronger predictor of PROMIS physical function and pain interference scores than pain intensity. Specifically, for physical function the overall R² ranged from 0.37 to 0.50 and for pain interference, R² ranged from 0.37-0.47. Moreover, NMAE ranged from 0.57 to 0.69 and 0.62 to 0.69 standard deviations for physical function and pain interference, respectively. In contrast, pain intensity models yielded R² ranging from 0.13-0.20. and NMAE ranged from 0.83 to 0.87. In contrast to ODI results, models involving PROMIS 1 W2 data did not perform better than those using ACT data. However, the combined sample did perform better than using PROMIS 1 W2 or ACT alone. Table 4 provides regression results for each model for use in translating scores from one scale to another, although the models predicting PROMIS pain intensity from RMDQ may not provide enough precision to be useful.

Table 4.

Model equation, R² and NMAE for regressions predicting PROMIS Physical Function, Pain Interference, and Average Pain Intensity scores from Roland-Morris Disability Questionnaire in PROMIS 1 W2 and ACT samples individually and combined.

PROMIS Score	Sample	Equation	R2	NMAE
Physical Function	PROMIS I W2 (n=179)	45.928 – 0.640 x RMDQ total score	0.37	0.69
	ACT (n=384)	51.928 – 0.869 x RMDQ total score	0.48	0.57
	ACT+PROMIS 1 W2 (n=563)	51.063 – 0.879 x RMDQ total score	0.50	0.62
Pain Interference	PROMIS I W2 (n=179)	54.762 + 0.707 x RMDQ total score	0.37	0.69
	ACT (n=384)	51.814 + 0.784 x RMDQ total score	0.45	0.64
	ACT+PROMIS 1 W2 (n=563)	52.154 + 0.811 x RMDQ total score	0.47	0.62
Average Pain	PROMIS I W2 (n=179)	4.746 + 0.129 x RMDQ total score	0.13	0.87
Intensity	ACT (n=384)	3.449 + 0.126 x RMDQ total score	0.13	0.89
	ACT+PROMIS 1 W2 (n=563)	3.510 + 0.161 x RMDQ total score	0.20	0.83

NOTE: NMAE = Normalized Mean Absolute Error; PROMIS I W2 = PROMIS 1 Wave 2 Pain and Depression study; ACT= Assessment of Chiropractic Treatment for Low Back Pain and Smoking Cessation in Military Active Duty Personnel. R² and NMAE are adjusted using linear equating to address regression to the mean.

Predicting ODI from PROMIS scales

The three PROMIS scores do reasonably well in predicting the ODI; Table 5 shows results of these models for CERC and PROMIS 1 W2 samples individually and combined. There is little variation in the precision of the prediction of ODI score according to sample, but the model using data from the combined sample yields the highest R² (0.68). Across all models, R² values ranged from 0.64 to 0.68 with NMAE ranging from 0.47 to 0.50. Beta coefficients for all three PROMIS scores were significant at p<.01.

Table 5.

Model equation, R², NMAE for regressions predicting Oswestry Disability Index from PROMIS Physical Function, Pain Interference and Pain Intensity for PROMIS I W2 and CERC samples individually and combined.

Sample	Equation	R2	NMAE
PROMIS I W2 (n=212)	−7.114 – 0.961 x PF + 1.134 x PI + 1.193 x Avg Pain	0.66	0.49
CERC (n=1524)	20.910 – 0.765 x PF + 0.541 x PI + 1.165 x Avg Pain	0.64	0.50
CERC + PROMIS I W2 (n=1736)	17.572 – 0.789 x PF + 0.623 x PI + 1.192 x Avg Pain	0.68	0.47

NOTE: NMAE = Normalized Mean Absolute Error; PROMIS I W2 = PROMIS 1 Wave 2 Pain and Depression study; CERC= Center for Excellence in Research for Complementary and Integrative Health. PF = PROMIS Physical Function T score; PI = PROMIS Pain Interference T score; Avg Pain = PROMIS Pain Intensity score. R² and NMAE are adjusted using linear equating to address regression to the mean.

Predicting RMDQ from PROMIS scales

Table 6 shows results of models predicting RMDQ from the three PROMIS scores. Beta coefficients for PROMIS physical function and pain interference scores across all models were significant at p<.05. However, the PROMIS pain intensity item was not a significant predictor in these models. Although, R² values for these RMDQ models (range = 0.46 – 0.58) are consistently lower relative to models predicting ODI (range = 0.64 – 0.68), they provide enough precision to be useful. As with ODI models, there is little variation in precision of the prediction of RMDQ score according to sample, but the pattern of results differs from that seen with the ODI: The results for the PROMIS 1 W2 sample yield the lowest precision, and the model using data from the combined samples does not improve the R² over the model that just includes the ACT sample.

Table 6.

Model equation, R², and NMAE for regressions predicting Roland-Morris Disability Questionnaire from PROMIS Physical Function, Pain Interference and Average Pain Intensity for PROMIS I W2 and ACT samples individually and combined.

Sample	Equation	R2	NMAE
PROMIS I W2 (n=179)	6.433 – 0.357 x PF + 0.289 x PI + 0.234 x Avg Pain	0.46	0.63
ACT (n=384)	5.451 – 0.356 x PF + 0.316 x PI + 0.140 x Avg Pain	0.58	0.56
ACT + PROMIS I W2 (n=563)	5.689 – 0.352 x PF + 0.307 x PI + 0.157 x Avg Pain	0.58	0.55

NOTE: NMAE = Normalized Mean Absolute Error; PROMIS I W2 = PROMIS 1 Wave 2 Pain and Depression study; ACT= Assessment of Chiropractic Treatment for Low Back Pain and Smoking Cessation in Military Active Duty Personnel. PF = PROMIS Physical Function T score; PI = PROMIS Pain Interference T score; Avg Pain = PROMIS Pain Intensity score. R² and NMAE are adjusted using linear equating to address regression to the mean.

Discussion

These analyses provide preliminary equations for translation of PROMIS physical function, pain interference, and pain intensity scores to and from two condition-specific instruments, the ODI and the RMDQ, using data from three large samples of patients with CLBP. Because relationships of PROMIS scores with the ODI and RMDQ were not strong enough to support latent variable equating, we used OLS regression with linear equating. The level of disability in the CERC and ACT datasets was relatively low compared to the PROMIS 1 W2 sample, so the combined results provide an increased range of disability. We focused on the three PROMIS scores that are most relevant to CLBP disability and recovery and are part of the ISS, but it is notable that the relationship of physical function and pain interference with ODI and RMDQ are stronger than that with pain intensity. This is especially true for the RMDQ.

Models reported here yielded useful coefficients to estimate PROMIS scores, particularly physical function and pain interference, from the ODI or RMDQ. Even so, the ODI and RMDQ total scores never accounted for more than 56% of the variance in PROMIS scores. This may be due in part to predicting a single PROMIS score (e.g., physical function) from multidimensional composites. We conducted exploratory analyses using individual items within the ODI and RMDQ as predictors of the PROMIS scores to determine whether use of items might improve the prediction. For ODI, we found that item level analyses increased the variance explained in the prediction models by 7% to 12%. In contrast, use of RMDQ items instead of total score increased the variance explained in the prediction models by between 1% and 2%. This discrepancy in impact is likely due to greater heterogeneity of ODI compared to RMDQ items.

In predicting PROMIS scores, ODI models are stronger than RMDQ models across all three PROMIS scores, and notably the addition of the more disabled PROMIS 1 W2 sample had more of an impact on precision of prediction equations for ODI than for RMDQ. In fact, the extended range of disability did not consistently improve predictions for RMDQ.

Models predicting ODI and RMDQ measure scores from PROMIS performed better than models predicting PROMIS from legacy measures, and this was particularly true for models predicting ODI. This better performance may be due to predicting a single score from three predictors as opposed to one.

Limitations

These analyses are based on relationships found in these datasets; the stronger relationships of PROMIS scores with ODI relative to RMDQ may be due to basic differences between the CERC and ACT samples. However, these differences are maintained in the PROMIS 1 W2 sample—thus, the superiority of ODI models are not explained by possible anomalies in the CERC and ACT samples. Another limitation is that correlations among observed scores precluded use of IRT equating. However, our best performing prediction equations achieved R² values of approximately 0.65 and higher, which correspond closely with accuracy reported by Schalet et al³⁸ in an IRT-based linking of the PROMIS Global Health physical summary scores with those from the VR-12. The mental health summary score linkages reported by Schalet et al were less accurate, corresponding closely with R² values of approximately 0.45, a level of accuracy that was reached or exceeded in many of the prediction models reported in this paper. The samples represent a limited range of disability, and results should be replicated in samples with a broader range of disability. Although it is unclear whether these prediction equations will be applicable to higher levels of disability than is represented here, a comparison of actual and predicted mean scores for quantiles of each dependent variable found prediction accuracy to be fairly stable across the range of scores in our data. Finally, although the addition of the PROMIS 1 W2 sample provides obvious benefits, the PROMIS scores from that sample are based on CAT administration of the PROMIS domains as opposed to the 4-item short forms in the PROMIS-29.

Conclusion

Researchers can use these results to translate PROMIS physical function, pain interference, and pain intensity scores to and from the ODI and RMDQ – the two condition-specific instruments most frequently used to evaluate interventions for CLBP. The empirical linkages can be used to facilitate comparisons across CLBP interventions and broaden interpretation of study results. Further studies are needed to evaluate the robustness of the prediction equations presented here.

Abbreviations:

ACT

Assessment of Chiropractic Treatment for Low Back Pain and Smoking Cessation in Military Active Duty Personnel

CAT

Computer Adaptive Test

CERC

Center for Excellence in Research for Complementary and Integrative Health

CLBP

Chronic Low Back Pain

CNP

Chronic Neck Pain

NIH

National Institutes of Health

NMAE

Normalized Mean Absolute Error

ODI

Owestry Disability Index

OLS

Ordinary Least Squares

PROM

Patient-Reported Outcome Measures

PROMIS

Patient-Reported Outcomes Measurement Information System

PROMIS-29

29-Item PROMIS Profile Measure

RMDQ

Roland-Morris Disability Questionnaire

SIP

Sickness Impact Profile