Correlation of patient reported outcomes among patients with chronic low back pain and controls

Abstract

Background

Chronic low back pain (CLBP) is a prevalent condition with significant physiological, psychological, social, and economic impacts. A range of patient-reported outcomes (PROs) are used to collect self-reported perceptions of patient health and well-being relating to the condition. Correlations between different types of PROs have previously been demonstrated – for example, between PROs that measure pain, and those that measure anxiety. Furthermore, PROs have evolved over time, and there exist strong correlations between more recently developed outcomes, such as PROMIS PROs, and legacy PROs. However, most studies in this area focus only on subjects with CLBP. It would be enlightening to determine whether the same correlations between PROS exist in healthy subjects, and whether and how the presence of CLBP moderates the relationships between these PROs. This comparative cross-sectional study hypothesizes that:

• • PRO metrics correlate with CLBP occurrence.

• • Legacy and PROMIS PROs are correlated in CLBP contexts.

• • CLBP moderates the relationships between PROs.

• • Latent factors may identify characteristics that most influence variance in outcomes.

Methods

We compared outcomes of legacy PROs with PROMIS PROs collected from participants aged 35–65 with (n = 133) and without (n = 100) CLBP. Welch t-tests compared PRO scores between groups. Linear regressions evaluated the relationship between legacy and PROMIS PROs, accounting for CLBP as a binary variable. An exploratory factor analysis identified latent factors summarizing variance in the PROs.

Results

Cases reported significantly lower scores than controls across all PROs except for activity level. Strong correlations emerged between several PROMIS metrics and two legacy PROs measuring pain intensity and disability. CLBP significantly moderated these relationships. Moderate correlations were noted between PROMIS metrics and pain catastrophizing and anxiety, with weaker correlations for activity level. Five latent factors were identified, capturing key characteristics that influence variance.

Conclusions

Legacy and PROMIS PROs performed similarly in terms of correlation with CLBP, suggesting they capture overlapping information. However, latent factor analysis indicates potential for designing more focused PROs, targeting characteristics in these factors, to better capture variance in outcomes across individuals with and without CLBP.

Clinical trial number

Not applicable.

Background

One of the most common health ailments, which has even been deemed “the nemesis of medicine” [1], is low back pain (LBP), affecting 50–80% of the general population [2–4]. LBP also has the potential to be very severe and has even been identified as the leading cause of disability worldwide [5, 6]. When LBP persists beyond or recurs after three months, it is characterized as chronic (CLBP) [7]. Approximately 10% of LBP cases will develop into CLBP [8], which has been associated with direct and indirect medical costs, reduced physical activity, decreased quality of life, and depression [9–12].

The biopsychosocial effects of CLBP on patients are often quantified using patient-reported outcomes (PROs). A range of PROs have been designed over the years to capture different aspects of patient perception of their health and well-being. In this study, 5 well-established (‘legacy’) PROs, and 5 more recently developed PROs (‘PROMIS’; see [13]), are utilized and compared, as described in [14]. Substantial research has been conducted that evaluates PROs among patients suffering from back pain (e.g [15, 16]), analyzes the change of PROs prior to and following intervention [17–22], identifies underlying factors to reduce the dimensionality of PROs [23, 24], and quantifies the correlation between different PROs among symptomatic patients [25–31]. However, it appears there are few current studies that investigate PRO correlations using data from subjects with and without CLBP (e.g [16]), and even less that considered the moderating effects of CLBP on those correlations. Patient perception of health is potentially modified by confounding biopsychosocial factors in the presence of CLBP [32], leading to significant changes in correlations between PROs for healthy subjects and those with CLBP. This study seeks to fill the current knowledge gap regarding how different aspects of a patient’s perception of their health and well-being (as quantified by PROs) relate to each other for individuals who suffer from CLBP, and how these relationships differ for individuals who experience no LBP.

Objective

The purpose of this comparative cross-sectional study, therefore, is to investigate how patient perceptions of different aspects of health and well-being (as captured by PROs) relate to one another in individuals with and without CLBP, and to determine how the presence of CLBP moderates these relationships. The study compares correlations between established (‘legacy’) PROs and newer Patient-Reported Outcomes Measurement Information System (PROMIS) [13] PROs and uses latent factor analysis to identify key underlying dimensions of patient reported outcomes.

This was done by performing the following analyses:

(1) Quantify the differences in PROs between the cases (study participants who suffer from CLBP) and the controls (participants who do not experience CLBP symptoms nor have more than ‘mild’ pain anywhere else in the body).

(2) Evaluate correlations between 5 relatively new PROMIS metrics and 5 more established (legacy) PROs (to determine whether they are measuring the same things) and investigate the moderating effects of CLBP on these correlations.

(3) Identify latent factors of patient perception of health and well-being (i.e., underlying dimensions which summarize the observed features) of the PROs data via an exploratory factor analysis (EFA) to identify the distinct characteristics captured by the PROs.

Hypotheses

This study hypothesizes that:

• PROs correlate with CLBP occurrence (i.e. there is a statistical difference in the distributions of PRO metrics gathered from controls vs. those gathered from people with CLBP).

• Legacy and PROMIS PROs that measure similar aspects of subject perception of health and well-being (such as pain, physical function or anxiety) are correlated within control and CLBP groups.

• The presence of CLBP moderates the relationships between legacy and PROMIS PROs, mentioned in the previous hypothesis.

• Latent factors may be used to identify characteristics that most influence variance in PRO distributions, potentially leading to a more succinct set of relevant PROs for people with CLBP.

Methods

Study design

This is a comparative cross-sectional study.

Setting

All data were collected within a university setting in Utah.

Participants

Ethics approval

This study was approved by the Institutional Review Board at Brigham Young University (ID IRB2023-008), and informed consent was provided by each participant prior to data collection. Our research was conducted in accordance with the Declaration of Helsinki [33]. Two cohorts of subjects were recruited for the study – individuals who suffer from CLBP (cases) and control subjects who do not experience CLBP symptoms (controls) [14]. Controls were recruited via flyers to the local community. Cases were recruited from individuals who were scheduled for physical therapy relating to CLBP. All participants were between the ages of 35 and 65 (which is the age range with the greatest CLBP prevalence [8]). The data used in this study were collected as part of a larger study examining movement patterns in individuals with CLBP compared to a non-LBP control group [14]. Data collection for this specific sub-study, which was determined a priori, took place between June 27, 2022, and April 12, 2023. For purposes beyond the scope of this work, but consistent with the larger study, it was required that participants be capable of standing and moving without an assistive device and be eligible for an MRI scan. Exclusion criteria for participants in the control group were current or history of treatment for lumbar spine pain, known scoliosis, and LBP pain intensity of greater than 2.5 on the NRS. This pain intensity restriction was implemented to ensure that our comparison group consisted of participants without significant pain, reducing the likelihood of confounding. For the CLBP group, the pain had to be primarily axial (as opposed to radiating, as seen in radiculopathy) and persistent for at least three months prior to the study. Additionally, participants in the CLBP group needed to have a pain intensity score of at least 2.5 on the NRS. This minimum pain intensity was chosen to account for potential measurement error, given that the minimal clinically important difference for patients with subacute or chronic low back pain is 2/10 on a visual analogue scale [34], Lastly, individuals with a history of knee or hip surgery were excluded from both, the case and control groups.

See also Fig. 1.

Fig. 1.

Individuals who were invited to participate in study and met eligibility criteria

Variables and measurement

The PROs of interest in this work include several commonly used and thoroughly validated questionnaires (referred to in this study as legacy PROs) and the relatively new PROMIS Short Forms metrics [13]. These data were collected as part of a larger study examining movement patterns in individuals with CLBP compared to a non-LBP control group [14], along with additional metrics. The rationale for selecting the particular PROs used in the study has been published in two protocol papers associated with that study [14, 35]. A brief description of each PRO is as follows:

• Pain Intensity: Low-back pain intensity was measured using a numeric rating scale (NRS) that ranges from 0 to 10 (0 representing ‘no pain’ and 10 representing the most severe pain possible). This scale is commonly used to measure pain intensity and has been well validated [36, 37].

• Oswestry-Disability Index (ODI): Spine-related disability was measured using the widely used ODI metric [38]. The ODI metric has been thoroughly vetted and is frequently used as the gold standard for quantifying disability related to LBP [26, 39] and has shown strong experimental reliability when evaluating patients with chronic pain [36].

• General Anxiety Disorder 2 (GAD): General anxiety disorder is an ailment in which an individual suffers from a generic worry. It is not caused by current stressful events in an individual’s life, though stress can aggravate the condition [40]. The GAD-2 questionnaire consists of two questions and is used to identify symptoms of anxiety over a two-week period, and has shown acceptable accuracy in past studies [41].

• Pain Catastrophizing Scale 6 (PCS): Pain catastrophizing is “an exaggerated negative orientation towards actual or anticipated pain experiences”, which past studies have found to be associated with higher pain intensities [42]. The PCS-6 is designed to capture three psychological components of catastrophizing: helplessness, rumination, and magnification [43].

• Short Form International Physical Activity Questionnaire (IPAQ): Self-assessed physical activity in the week prior to subject participation in the study was estimated in terms of Metabolic Equivalent Task (MET) minutes per week. Previous studies have compared IPAQ scores to objective metrics of physical activity, and found it to be moderately reliable for estimating vigorous physical activity [44] but an uncertain metric for evaluating total aerobic activity [44, 45]. However, due to the innately subjective nature of patient-satisfaction, an individual’s self-perception of physical activity may be more valuable than precise estimations of activity [36].

• Patient-reported outcomes measurement information systems (PROMIS) metrics: PROMIS scores are “a set of person-centered measures that evaluates and monitors physical, mental, and social health in adults and children” [46]. Past studies have evaluated PROMIS metrics and concluded them to be “efficient, flexible, and precise” [13]. Questionnaire responses for each PROMIS metric are used to calculate T-scores, which are intended to achieve a normal distribution with a mean metric score of 50 and a standard deviation of 10 for the general population [47, 48]. Past studies have found that PROMIS metrics T-scores are less prone to floor effect, easier to implement in clinical settings, and more comprehensive than other metrics such as the ODI [26, 28]. The short-form PROMIS metrics implemented in this research included the PROMIS Pain Interference 4a, PROMIS Physical Function 6b, PROMIS Sleep Disturbance 6a, PROMIS 4-item Depression, and PROMIS 4-item Anxiety.

All demographic data and PROs were collected using an electronic survey sent to study participants (the IPAQ questionnaire was included in the study shortly after it was initiated, hence the IPAQ MET Minutes /Week scores from the first 30 participants are missing from this study). The participant responses were de-identified, then uploaded to the Back Pain Consortium Data Portal [14], along with other features of interest, to be used in ongoing and future research initiatives [35].

Data processing

It has been observed that IPAQ MET scores are prone to overestimations [49]. Past studies have warned that including extreme outliers in a dataset can lead to spurious findings [50], so can the removal of outliers without due cause and documentation [51]. To mitigate both risks, subsequent statistical tests including the IPAQ MET scores were conducted twice. The first analyses were conducted after filtering extreme outliers in the IPAQ responses (i.e., z-scores > 3, which is slightly more aggressive than the recommendation by Tabachnik and Fidell [52]). The second analysis was conducted including the entire dataset. The effects of including the entire dataset in the analyses are reported in the results, with periodic comments on the impact of removing outliers.

It was noted that the ODI and IPAQ MET Minutes/Week scores were both right-skewed. This was mitigated with a square-root transformation (in both cases, this was the minimal transformation needed to achieve a Fisher-Pearson coefficient within the range of -1 to 1 while maintaining as much of the original data characteristics as possible).

Statistical methods

Several of the analyses performed in this study assume the data have a normal distribution. While the PROMIS metrics are designed to achieve a normal distribution [47, 48], the other PROs of interest in this study were not designed with the intention of achieving any particular spread. A fair statistical comparison between the PRO distributions requires that any distributions that are significantly skewed away from normal, be transformed to bring them within an acceptable range of normality. The skewness of the PRO distributions was quantified using Python’s skew function to calculate the Fisher-Pearson coefficient. Significant skewness (designated in this study as having a Fisher-Pearson coefficient outside the bounds of -1 to 1, which indicates right- and left-skewness respectively) was mitigated by transforming the data prior to conducting statistical analysis. An example transformation is to raise the data values to an appropriate power; in this study, taking the square root of two of the distributions was adequate to remove unwanted skewness, as discussed below.

Several statistical analyses were conducted to achieve the aims described in the Introduction and are described below. In each test, a p-value of less than 0.05 was considered statistically significant, after accounting for multiple comparisons using the Holm correction factor.

Aim one: comparison of PROs from cases and controls

Welch T-Tests were used to evaluate the differences of the PROs distributions between the cases and controls. The p-value of the comparison of Pain Intensity between the two groups was not calculated (calculating this statistic would have been improper due to the inclusion-exclusion criteria – individuals with a Pain Intensity below 2.5 on the NRS were excluded from the cases, and individuals with a Pain Intensity above this value were excluded from controls, as mentioned above in the data collection description. Cohen’s d values were calculated to determine the magnitude of effect sizes.

Aim two: correlations between PROs

The second objective of this study was to evaluate the strength of the correlation between legacy PROs (Pain Intensity, ODI, GAD, PCS-6, and IPAQ MET Minutes/Week) with the PROMIS metrics (Pain Interference, Physical Function, Sleep Disturbance, Depression, Anxiety), and test whether CLBP is a moderating factor in the relationships. Linear regressions were employed to quantify the strength and statistical significance of the correlations (after skewness has been corrected for, when necessary). The linear regression models were of the following forms:

1

2

3

4

In each equation, represents one of the five legacy PROs and represents one of the five PROMIS metrics, for a total of 25 regressions analyses performed.

• In Eq. 1, the simplest form of linear regression is considered. The coefficient represents the y-intercept and the coefficient represents the slope depicting the average effect on the dependent variable per unit increase in the independent variable . In this model, the relationship between the legacy and PROMIS PRO is independent of the presence of CLBP.

• In Eq. 2, the additional coefficient represents the difference in y-intercepts between cases and controls. CLBP status is indicated by , which has a value of 1 for subjects who suffer from CLBP and a value of 0 for control subjects. The nature of the relationship between PROs (i.e., the slope of the regression) in this model is the same for both cases and controls, but the presence of CLBP is associated with an offset of the dependent variable.

• In Eq. 3, the coefficient represents the difference in slope for the y-intercept between cases and controls, indicating that the nature of the relationship between PROs (i.e., the slope) is altered by the presence of CLBP and therefore differs for the two groups.

• In Eq. 4, the dependent variable is estimated using the average value for the two groups. This model is applicable when the average value of the legacy PRO () differs, but the legacy PRO and PROMIS score () are not related to each other.

To reduce the risk of Type I Error, each additional model feature was added incrementally after the base model was proven statistically significant. The most appropriate equation to model the relationship between the PROMIS metric, legacy PRO, and the effect of CLBP was determined using the logic displayed in the decision tree in Fig. 2. At each decision tree node, the F-statistic and corresponding p-value was calculated to determine the statistical significance of the additional model feature (Eq. 5).

Fig. 2.

Decision tree indicating the process for selecting whether a relationship between the legacy and PROMIS PRO exists, and if so in what manner the presence of CLBP influences the relationship

5

– The variance explained by the model with features.

– The variance explained by the model with features.

, – The number of variables in the regression models being compared (Eq. 1 and Eq. 4 have one variable, Eq. 2 has two variables, and Eq. 3 has three variables). When testing the statistical significance of Eq. 1 and Eq. 4, = = 0.

– The number of datapoints observed.

The strength of the correlations was evaluated using the Cohen cut points for the absolute values of the Pearson’s correlation coefficients (=0.10–0.30 is considered small, =0.31–0.50 is considered moderate, and >0.50 is considered large [49, 53]).

Aim 3

Identify latent factors of patient perception of health and well-being.

The third aim of this analysis was to uncover the minimal set of independent theoretical constructs that could account for the observed trends in the PRO data (including both the legacy and PROMIS metrics). These constructs are comprised of latent factors, which are unobservable independent variables that are either causally related to or associated with the observed variables. Latent factors are essentially linearly independent hidden components that summarize the variance in the raw data. In contrast, however, to the original PROs, there is no covariance between latent factors. We utilized a machine learning algorithm to identify these latent factors through an EFA [54].

Prior to analysis, the z-scores of each observed PRO were calculated so that each would have an equal weight in the EFA. The PROMIS metric z-scores were calculated according to the theoretical distribution by subtracting 50 and dividing by 10 [48]. Other PRO z-scores were calculated using Python’s z-score function (after mitigating skewness when necessary).

The latent factors were identified by conducting a principal component analysis on the z-score data and applying a varimax rotation, which maximizes the variance explained by each original component by a single factor. This process was repeated multiple times to determine how many resultant latent factors were needed to capture at least 50% the variance in each of the original z-score data. I.e., the first EFA analysis identified a single latent factor, the second EFA analysis identified two latent factors, etc., until the resultant latent factors identified could account for at least 50% of the variance in every PRO z-score data.

The Coefficients of Determination were calculated between the PRO z-scores and the resultant latent factors to provide an interpretation of the factors.

Results

Participants

A total of 233 subjects were recruited for the study – 133 cases, 100 controls (see [26] for a study of similar size involving PROs). Participant demographics are summarized in Table 1.

Table 1.

Summary of participant demographics, in terms of the average values and standard deviation (SD)

Demographic	Control participants	Participants with CLBP
Number of subjects	100	133
Age [SD] (yrs)	47.8 [7.5]	49.0 [8.8]
Height [SD] (cm)	174.3 [9.5]	172.2 [18.0]
Weight [SD] (kg)	79.5 [15.9]	85.2 [20.2]
BMI [SD] (kg/m2)	26.1 [4.5]	28.3 [6.1]
Sex at Birth (Female / Male)	43 / 57	64 / 69

Differences in PRO between cases and controls

Boxplots were used to visually depict the differences in the PRO scores between cases and controls in Fig. 3. A summary of the results from the Welch T-tests and Cohen’s d effect sizes are recorded in Table 2.

Fig. 3.

Boxplots contrasting the PRO values between controls (blue) and cases (red). The raw data collected (prior to the transformation to achieve normal sample distributions) are depicted here

Table 2.

Summary of the Welch T-tests to evaluate the difference in PRO among the control and case groups (group averages and standard deviations are reported in terms of the Raw data, and the t-statistic and corresponding p-values are reported in terms of the normalized dataset when necessary to satisfy the assumptions of a Welch T-test); Cohen’s d for effect size

PRO	Control Cohort Mean [SD]	Cohort w/ CLBP Mean [SD]	t-statistic	Adjusted p-value	Cohen’s d
Pain Intensity	0.59 [0.74]	5.45 [1.52]	-30.61	N/A**	4.07
ODI	3.57 [4.83]	27.41 [15.32]	-19.85	0.000*	2.10
GAD	0.63 [1.04]	1.56 [1.75]	-5.01	0.000*	0.65
PCS-6	5.54 [4.53]	9.50 [5.79]	-5.85	0.000*	0.76
IPAQ MET Minutes / Week	5648 [5758]	4678 [5958]	1.72	0.087	-0.17
PROMIS Pain Interference	44.84 [5.71]	61.56 [6.99]	-20.07	0.000*	2.62
PROMIS Physical Function	55.34 [5.61]	41.29 [7.10]	16.91	0.000*	-2.20
PROMIS Sleep Deprivation	46.12 [7.24]	55.78 [8.36]	-9.43	0.000*	1.24
PROMIS Depression	45.86 [6.89]	42.46 [9.45]	-6.17	0.000*	-0.41
PROMIS Anxiety	47.16 [7.09]	51.98 [9.44]	-4.45	0.000*	0.58

* Statistical significance was observed in the comparison

** The p-value of the Pain Intensity comparison was not calculated due to the threshold criterion involving this metric

PRO correlations

A summary of the Fisher-Pearson coefficients for each PRO investigated in this study (prior to and post transformation) are recorded in Table 3. The resulting correlations between legacy and PROMIS PROs were generated and depicted as shown in Fig. 4. The types of relationship present are categorized as:

Table 3.

Summary of the data collected and the normality skewness before and after transformation (~ in the after-transformation column indicates that no transformation was applied, and the Fisher-Pearson coefficient remained constant)

PRO	Fisher-Pearson (before transformation)	Transformation Applied	Fisher-Pearson (after transformation)
Pain Intensity	0.178	None	~
ODI	1.005	Square-root	0.001
IPAQ MET Minutes / Week	3.106	Square-root	0.685
PROMIS Pain Interference T-Score	0.148	None	~
PROMIS Physical Function T-Score	-0.095	None	~
PROMIS Sleep T-Score	0.189	None	~
PROMIS Depression T-Score	0.726	None	~
PROMIS Anxiety T-Score	0.599	None	~

Fig. 4.

Scatterplots and the respective line of best fit (when applicable) depicting the relationship between the different PROs for subjects who experience CLBP symptoms (red) and asymptomatic subjects (blue)

1. A single relationship is present for both cases and controls (a single purple line, described by Eq. 1);

2. A relationship of the same type (i.e. the same slope) is present for both cases and controls, but with an offset between them (two purple lines, as given by Eq. 2);

3. Two different relationships exist for cases and controls (indicated by a purple and a blue line, as described by Eq. 3).

In both types 2 and 3, the presence of CLBP modifies the relationship between the legacy and PROMIS outcomes. In type 2, the relationship is the same but experiences an offset when CLBP is present; in type 3, the relationship changes in the presence of CLBP. If the relationship changes, it can be concluded that the legacy and PROMIS PROs that are being compared are not equivalent but contain underlying characteristics that are modified in different ways by the presence of CLBP. The different model parameter values (i.e. the constants required by the relevant equation(s)), adjusted p-value associated with the given relationship, and the absolute value of the Pearson’s correlation coefficient () from the regressions are summarized in Table 4.

Table 4.

Summary of the model parameters and overall p-values and values for each PRO combination correlation. Note that the interaction terms ( and ) indicate the difference in the y-intercept and slope from the baseline values ( and ). If no statistically significant model was found to depict the relationship, the statistically insignificant parameters from eq. 1 were reported

Legacy PRO	PROMIS T Score					Adjusted p-value	r
Pain Intensity	Pain Interference	-1.874	0.055	-0.367	0.070	0.000	0.913***
	Physical Function	2.218	-0.029	7.890	-0.083	0.000	-0.906***
	Sleep Deprivation	-1.102	0.037	4.513		0.000	0.88***
	Depression	-0.505	0.024	4.709		0.000	0.881***
	Anxiety	-1.005	0.034	4.704		0.000	0.884***
ODI	Pain Interference	-5.815	0.160	1.009		0.000	0.910***
	Physical Function	10.124	-0.158	1.456		0.000	-0.909***
	Sleep	-1.434	0.060	0.524	0.046	0.000	0.852***
	Depression	-1.863	0.070	3.224		0.000	0.832***
	Anxiety	-1.957	0.070	3.343		0.000	0.833***
PCS	Pain Interference	-7.073	0.273			0.000	0.510***
	Physical Function	22.413	-0.309			0.000	-0.523***
	Sleep Deprivation	-8.062	0.307			0.000	0.503***
	Depression	-9.434	0.327	1.798		0.000	0.600***
	Anxiety	-10.759	0.346	2.290		0.000	0.627***
GAD	Pain Interference	-0.910	0.030			0.000	0.405**
	Physical Function	2.121	-0.029			0.000	-0.354**
	Sleep Deprivation	-1.179	0.037			0.000	0.436**
	Depression	-2.172	0.059			0.000	0.675***
	Anxiety	-2.803	0.069	0.148		0.000	0.808***
IPAQ MET Minutes / Week	Pain Interference	107.843	-0.860			0.003	-0.222*
	Physical Function	5.770	1.170			0.000	0.276*
	Sleep Deprivation	127.155	-1.284			0.000	-0.283*
	Depression	111.710	-1.021			0.003	-0.232*
	Anxiety	99.133	-0.764			0.018	-0.167*

* small correlation strength ( is between 0.10–0.30)

** moderate correlation strength ( is between 0.30–0.50)

*** large correlation strength ( is between 0.50–1.00)

Latent factors

A set of latent factors that accounted for a simple majority (over 50%) of the variance in the data was identified by EFA analysis. A total of five latent factors were required to capture this level of variance in the original PRO z-score data. The Pearson’s correlation coefficients () between the latent factors and PROs were calculated. The first factor demonstrated a strong correlation with Pain Intensity, ODI, PROMIS Pain T-score, and a strong negative correlation with the PROMIS Physical Function T-score. Due to its strong correlation with PROs that describe an individual’s physiological limitations, this latent factor was labeled “Pain and Physical Limitations”. The second factor identified demonstrated a strong correlation with PROMIS Depression T-score, PROMIS Anxiety T-score, and the General Anxiety Disorder metric, and was consequently labeled “Psychological Distress”. The final three latent factors nearly replicated the PROMIS Sleep Deprivation, IPAQ MET Minutes/Week metrics, and Pain Catastrophizing-6 scores. These were subsequently named “Sleep Deprivation”, “Physical Activity”, and “Pain Catastrophizing”. See Fig. 5a; Table 5 for further details. In order to interpret the Pearson correlation statistics in Table 5, it should be noted that a value of + 1 means that the data are perfectly linearly correlated, a value of -1 indicates that they are perfectly negatively linearly correlated, and that a value of 0 means that they are uncorrelated. The correlations reported here coincide with the correlation of determinations observed between the original features (see Fig. 5b).

Fig. 5.

(a) The correlation of determination between the five factors identified -Physiological Ailment, Psychological Ailment, Sleep Deprivation, Physical Activity, and Pain Catastrophizing- with the original features PROs of interest and (b) and the correlation of determination between the PROs of interest

Table 5.

Summary of the Pearson’s correlation / p-values between the original pros of interest and the identified latent factors

PROs	Pain and Physical Limitations	Psychological Distress	Physical Activity	Sleep Deprivation	Pain Catastrophizing
IPAQ MET Minutes / Week	-0.101 / 0.118	-0.058 / 0.368	0.984 / 0.000	-0.075 / 0.248	-0.063 / 0.329
Pain Intensity	0.904 / 0.000	0.140 / 0.030	0.009 / 0.888	0.116 / 0.073	0.084 / 0.194
ODI	0.877 / 0.000	0.264 / 0.000	-0.074 / 0.250	0.223 / 0.001	0.145 / 0.025
PROMIS Pain Interference	0.907 / 0.000	0.229 / 0.000	-0.072 / 0.268	0.140 / 0.031	0.132 / 0.041
PROMIS Physical Function	-0.880 / 0.000	-0.185 / 0.004	0.139 / 0.032	-0.169 / 0.009	-0.171 / 0.008
PROMIS Sleep Deprivation	0.449 / 0.000	0.245 / 0.000	-0.115 / 0.075	0.838 / 0.000	0.142 / 0.028
PROMIS Depression	0.282 / 0.000	0.792 / 0.000	-0.148 / 0.022	0.119 / 0.067	0.180 / 0.005
PROMIS Anxiety	0.181 / 0.005	0.894 / 0.000	-0.065 / 0.318	0.059 / 0.367	0.201 / 0.002
GAD	0.167 / 0.010	0.899 / 0.000	0.065 / 0.315	0.144 / 0.026	0.075 / 0.247
PCS	0.315 / 0.000	0.404 / 0.000	-0.094 / 0.148	0.142 / 0.028	0.839 / 0.000

Discussion

Comparison of PROs from cases and controls to show impact of CLPB on outcomes

For the t-statistic reported in Table 2, the degrees of freedom are given by 133 + 100-2 (n1 + n2-2, where n1 and n2 are the number of samples); then, from a standard table, assuming 95% confidence is required, the absolute value should be above 1.97 in order for the relationship to be statistically significant. It was observed that individuals who reported CLBP symptoms also reported PROs that reflect a statistically significant lower quality of life in terms of ODI, GAD, PCS, PROMIS Pain Interference, PROMIS Physical Function, PROMIS Depression, and PROMIS Anxiety. A stark difference in the Pain Intensity scores between the two cohorts was also observed. These findings are consistent with results from past studies that employed similar metrics [15, 16].

There was no statistically significant difference in the IPAQ MET Minutes/Week scores of the cases and controls; this was the case when considering the entire dataset and after removing outliers. Several previous reported works have evaluated the role of CLBP on self-reported or measured physical activity, sometimes with conflicting results. For example, previous articles have reported a similar conclusion that CLBP may either prompt normal or increased physical activity as subjects ignore the pain [55] or reduced physical activity consequent to fear avoidance [56, 57], depending upon other psychosocial factors. Consequently, there may exist sub-groups of patients with CLBP who exhibit reduced, altered, or increased physical activity, and the contrasting effects of the different sub-groups appear to negate each other when evaluating patients with CLBP as a whole [58, 59].

The effect sizes derived from the comparison between individuals with CLBP and the control cohort (the final column in Table 2) reveal several key insights and confirm the above: the most pronounced differences were observed in pain-related measures, with Pain Intensity (d = 4.07), PROMIS Pain Interference (d = 2.62), and ODI (d = 2.10) demonstrating extremely large effects, consistent with the expected functional and perceived impact of chronic pain conditions. Measures of physical functioning also showed substantial differences. PROMIS Physical Function (d = -2.20) indicated a significant decline in self-reported function among those with CLBP, while PROMIS Sleep Deprivation (d = 1.24) pointed to a large effect on sleep quality, a well-documented comorbidity of chronic pain.

Psychological measures yielded more moderate effect sizes: PCS-6 (d = 0.76) and GAD (d = 0.65) suggest a moderate increase in pain catastrophizing and anxiety symptoms, respectively, among participants with CLBP. PROMIS Anxiety (d = 0.58) and PROMIS Depression (d = -0.41) also reflect moderate impacts, though the negative sign in the depression score suggests slightly lower levels of self-reported depression in the CLBP group—a finding that may warrant further investigation, especially in light of possible ceiling effects or self-report biases. Confirming the findings above, IPAQ MET Minutes (d = -0.17) showed only a small, nonsignificant difference in self-reported physical activity, which may reflect compensatory behavior or variability in how activity is perceived and reported in pain populations.

Relationships between PROs

Unsurprisingly, there is a strong relationship between legacy PROs that measure pain and physical function (Pain Intensity and ODI) and PROMIS metrics that measure similar characteristics (PROMIS Pain Interference and Physical Functionality). Moderate to strong correlations were also observed between the legacy metrics that evaluate psychological well-being with the PROMIS metrics. These findings are consistent with observations from past investigations [26, 31, 39, 42, 60]. In 12 out of 25 cases, the relationship between PROs remained the same in the presence or absence of CLBP. For 10 of the relationships, there was a vertical shift in the data for cases, compared with controls, but the slope remained the same. This indicates that the PROs being compared had different underlying characteristics that were modified by CLBP. For 3 of the comparisons, the relationship between pairs of PROs was modified, both in terms of vertical shift and slope, indicating that they embodied characteristics that were significantly changed by the presence of CLBP.

Specific observations regarding the relationships between PROs include the following:

PROMIS metrics with Pain Intensity: Large positive correlation strengths were observed between Pain Intensity with PROMIS Pain Interference, Sleep Depression, Anxiety, and Depression for both cohorts, and a negative correlation with PROMIS Physical Function. The presence of CLBP altered the y-intercept in each of these cases – individuals with CLBP reported higher pain intensities for any average PROMIS score, e.g., individual with higher PROMIS Sleep Deprivation scores reported higher Pain Intensity scores for both cohorts, but at any given Sleep Deprivation score, the cases reported higher Pain Intensities than their control counterparts. Furthermore, the presence of CLBP had an interactive effect on the slope of the correlation for the cases of Pain Interference and Physical Function: the slopes of the relationships were steeper for cases than controls. This indicates that on average, a step increase of Pain Interference and Physical Function reported a more dramatic change in pain intensity for cases than controls.

PROMIS metrics with ODI

Similar trends were observed between the PROMIS metrics and ODI scores. Increases in Pain Interference, Sleep Deprivation, Anxiety, and Depression as measured by the PROMIS metrics were associated with increases in disability as measured by ODI. Increases in PROMIS Physical Function were associated with decreases in ODI. The presence of CLBP significantly moderated the y-intercept in each of these relationships. Cases had higher ODI scores on average than controls who reported similar PROMIS metrics. Furthermore, the presence of CLBP also had an interactive effect on the slope of the relationship between PROMIS Sleep Deprivation and ODI – increases in Sleep Deprivation scores were associated with a more dramatic increase in ODI scores for cases than controls.

PROMIS metrics with PCS

Large correlation strengths were observed between the Pain Catastrophizing metrics with the PROMIS metrics. For the PROMIS metrics that evaluate Pain Interference, Physical Function, and Sleep Deprivation, the presence of CLBP was not a significant moderating factor in the relationship; e.g., the relationship between Pain Interference and Pain Catastrophizing followed the same trend for both cohorts. However, CLBP was a significant moderating factor in the correlations for PROMIS Depression and PROMIS Anxiety. In both relationships, the presence of CLBP was associated with a higher y-intercept. Increases in PROMIS Depression were associated with increases in Pain Catastrophizing for both cases and controls, but for a given PROMIS Depression score, cases on average had a higher Pain Catastrophizing score than controls.

PROMIS metrics with GAD

Moderate to large correlation strengths were observed between General Anxiety with the PROMIS metrics. The presence of CLBP was considered a moderating factor only in the relationship with PROMIS Anxiety, in which cases had a higher y-intercept than controls.

PROMIS metrics with IPAQ minutes/week

Statistically significant but weak correlations were observed between the PROMIS metrics and IPAQ MET Minutes/Week ratings. IPAQ MET Minutes/Week demonstrated a negative correlation with Pain Interference, Sleep Deprivation, Depression, and Anxiety. The PRO displays a moderate positive correlation with Physical Function. These observations were consistent both when outliers were removed from the dataset and when all observations were included. Past studies have also found physiological and psychological benefits to physical activity [61–64]. Future studies may find stronger correlations between physical activity and reported outcomes by using objective activity measurements rather than self-reported activity [44, 45].

Distinct components of patient perception of health and well-being

Five latent factors were extracted that best capture the variance in the PRO data across the entire dataset. The characteristics of these combined outcomes were labeled: “Pain and Physical Limitations”, “Psychological Distress”, “Sleep Deprivation”, “Physical Activity”, and “Pain Catastrophizing”. The latter three were similar in nature to three existing PROs, but the first two combine aspects of more than one of the studied PROs and might inspire modified PROs in the future that more efficiently capture changes in patient’s perception of their health and well-being in the presence of CLBP.

Limitations

While this study was strengthened by its relative large sample size and by the inclusion of individuals with and without CLBP there were also some limitations. All participants were recruited from a single institution within a specific geographic region, which may limit the generalizability of the findings to broader or more diverse populations. This homogeneity in the sample may not capture regional, institutional, or cultural differences that could influence the outcomes. In addition, due to CLBP being a very heterogeneous ailment, caution should be taken before generalizing results to a population [65, 66].

Another limitation is that the subjects were recruited as part of a larger, ongoing study. Inclusion criteria for the study required that the subjects be eligible to receive an MRI scan and perform a variety of functional motions [67]. This may have deterred individuals with metal implants, more severe CLBP symptoms, or higher degrees of kinesiophobia [68–70] from participating.

One inherent limitation of any PRO is that all self-assessments are subjective and thus prone to biases: e.g., past investigations have observed that self-reported physical activity is generally higher than objective activity measurements [49], and self-reported pain intensity is associated with direct observations of pain behavior but influenced by several moderating factors [71]. However, many studies consider individual’s self-perception as a vital (and sometimes superior) metric of back pain due to its highly subjective nature [9, 36, 72].

Conclusions

Data from PRO questionnaires submitted by 233 participants (133 cases of CLBP and 100 controls) were evaluated - five ‘legacy’ PROs and five more modern PROMIS PROs, all commonly used for patients with CLBP. Our hypothesis that the PROs correlate with the presence of CLBP was supported. One exception was the IPAQ MET Minutes/Week outcome, which was not associated with CLBP. This indicates that people with and without CLBP have similar activity levels. This finding is consistent with reports from past studies [55, 73]. The strongest correlations with CLBP were present for PROs relating to pain and physical function, with the legacy and PROMIS PROs performing similarly in each category. This finding is also consistent with our hypothesis.

An overall outcome of these observations is that either the legacy or PROMIS PROs contain similar information, but the PROMIS PROs are preferred for the normal distribution of data obtained (as opposed to the skewed distributions present for certain legacy PROs).

The change in relationships between PROs for individuals with and without CLBP indicates that when applying these PROs in broader populations to investigate CLBP, it is important to account for a systematic shift in response patterns among individuals with CLBP. The lack of a consistent continuum in outcomes suggests that interpretation of these results must be adjusted accordingly. In this context, CLBP functions as a confounding variable that may bias associations unless appropriately controlled for in analysis.

The identification of latent factors from the PRO data suggests that integrating elements from multiple existing PROs may enable the development of a more targeted instrument that more accurately captures the specific impact of CLBP on patients’ experiences and functioning.

Author contributions

SB, AB, UM and DF contributed to the conception of the work; UM, KC, AG and AG made substantial contributions to data acquisition; SB wrote the main manuscript; SB, AB, UM and DF interpreted the data; SB prepared all figures. All authors reviewed the manuscript.

Funding

Research reported in this publication was supported by the NIH HEAL Initiative under award number NIAMS UH3AR076723. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Data availability

Data are available from the authors upon reasonable request and with permission from the NIH HEAL Initiative.

Declarations

Ethics approval and consent to participate

This study was approved by the Institutional Review Board at Brigham Young University (ID IRB2023-008), and informed consent was provided by each participant prior to data collection. Our research was conducted in accordance with the Declaration of Helsinki.

Competing interests

The authors declare no competing interests.