The use of registry data to assess clinical hunches: An example from the Swedish quality registry for pain rehabilitation

1 Introduction

Experienced health care professionals sometimes “feel” that things are in a certain way. While it is important to listen to “clinical hunches” emitted by experienced clinicians, it is also essential to critically assess them. If possible, such impressions should be tested in a study. In the present study, the clinical impression of health professionals in a specialized pain centre was tested by using data from questionnaires answered by patients.

Chronic pain is associated with significant distress in the form of anxiety, depression, dysfunctional coping behaviours, prolonged sick leave, low participation in social activities, and/or unresponsiveness to routine pharmacological treatments [1,2,3,4]. The difficulty and often impossibility of “curing” chronic pain results in many frustrated and desperate patients [5]. Based on data from the Swedish quality registry for pain rehabilitation (SQRP) answered by chronic pain patients assessed at a multidisciplinary pain centre, and using hierarchical cluster analysis, it was possible to define four subgroups of chronic pain patients [6]. One of the groups was characterized by high “psychological strain” and by the most negative situation with respect to pain characteristics (intensity and spreading). For instance, in this high “psychological strain” group, which represented 17% of the study population, the 25th percentile value of the anxiety and depression subscales of hospital anxiety and depression scale (HAD) was 11 points in both cases – for each of the HAD subscales, a value of score of 0–7 is considered a non-case, 8–10 is a doubtful case, and 11–21 indicates a case [6]. Hence, 75% of the patients in this group reported substantial levels of anxiety and/or depression. Chronic pain patients are sometimes referred to as being “difficult” [7].

During and after the recent pandemic, experienced pain physicians and other health professionals such as physiotherapists and psychologists at the Pain and Rehabilitation Centre (PRC), Linköping University Hospital, Linköping, Sweden, have been contending that PRC patients have gradually become more complex and “difficult” over time. The time frame has not been specified but, on the one hand, it is longer than only a few years and, on the other hand, can at most refer to perhaps a maximum of 15 years. This clinical impression has not been substantiated by data. The aim of this repeated cross-sectional registry-based study was to examine this clinical claim by using local data retrieved from SQRP between 2009 and 2022.

2 Methods

The SQRP records patient-reported outcome measures (PROMs) from a majority of specialist chronic pain units/departments in Sweden [8]. PROMs are completed by patients on up to three occasions: before the first visit (baseline assessment) and for those who participate in interdisciplinary pain rehabilitation programs (IPRP), immediately after IPRP, and at a 12-month follow-up. In this article, only baseline data from PRC patients were used. The PROMs capture a patient’s background, pain intensity, pain-related cognitions, psychological distress symptoms, as well as activity/participation aspects and health-related quality of life variables. The questionnaires available in the SQRP have been amply described in other publications such as, for instance, Bäckryd et al., Ringqvist et al., and Alföldi et al. [6,9,10], and will therefore not be described in detail here.

In 2016, some questionnaires were added to the SQRP, and the analysis in the present study was therefore made in two stages. First, data from 2009 to 2022 were explored with a more restricted set of PROMs available during these 14 years. To this effect, three time periods were defined: period 1 (2009–2012), period 2 (2013–2016), and period 3 (2017–2022). Then, in a second step, data from 2016 to 2022 were explored with a higher number of PROMs, focusing on the dichotomy between pre-pandemic (2016–2019) and pandemic time (2020–2022). PROM abbreviations are shown in Table 1.

As in multiple previous publications such as, for instance, Bäckryd et al. and Skogberg et al. [6,11], multivariate data analysis by projection was used to explore the correlation structure of PROMs with SIMCA version 16, Sartorius Stedim Biotech, Umeå, Sweden [12,13]. Principal component analysis (PCA) and (orthogonal) projections to latent structures-discriminant analysis ((O)PLS-DA) models were computed. Briefly, PCA is an unsupervised technique that models the correlation structure of a dataset and thereby enables identification of multivariate outliers, as described in more detail elsewhere [14]. Here, the first principal component (PC1) was used as a multivariate “summary” measure for the complexity of the patient. Moreover, (O)PLS-DA, which is a supervised technique, was used for group comparisons, enabling the identification of the X-variables (i.e., predictors) most responsible for group discrimination while at the same time taking the whole correlation structure of the material into consideration. R ² describes the goodness of fit and Q ² describes the goodness of prediction. For univariate and bivariate analyses, IBM SPSS Statistics (version 28.0; IBM Corporation, Route 100 Somers, New York, USA) was used. Non-parametric statistics such as the Kruskal–Wallis test and Spearman’s rho for bivariate correlation were used as appropriate, as well as the chi-squared test. A p-value of <0.05 was considered significant.

3 Results

3.1 First step: Analysis of patients 2009–2022

A total of 24 PROMs in 8,133 patients were available (Table 1). After exclusion of patients having >50% missing values as well as both strong outliers and serious moderate outliers, the final PCA model had n = 7,563, 3 PCs, R ² = 0.48, and Q ² = 0.29. PC1 captured about 30% of the variation. The score plot and loading plot are shown in Figures 1 and 2, respectively. The two plots are complementary and must be interpreted together. For instance, patients on the right-hand side of Figure 1 report higher-than-average levels of pain intensity (see the NRS7d symbol in Figure 2). Likewise, patients on the left-hand side of Figure 1 report higher-than-average quality of life (i.e., high values on the EQ5D-index, see Figure 2). As can be seen in Figure 2, age and gender are not important in the model (corresponding dots are near the origin of the plot). PC1 did not change statistically over time (Figure 3, p = 0.177). Moreover, there was no correlation between time (i.e., the year) and PC1 (rho = −0.014; p = 0.21). Moreover, a PLS-DA model with the group belonging to the Y-variable (three groups) was computed. The model was very weak with 1 predictive PC, R ² = 0.048, and Q ² = 0.045. Hence, the goodness of fit was less than 5%, i.e., there was almost no difference between the three groups. The proportion of strong outliers (n = 57) did not differ between the first half of the study period (2009–2015) and the second half of the study period (2016–2022) (0.7 vs 0.9%, respectively; p = 0.282).

Figure 1

Score plot showing the first and second principal components (t[1]) and t[2], respectively) on patient-reported outcomes of 7,563 chronic pain patients. Each dot represents a patient. The similarity between the three time periods is illustrated by the colour overlap, most blue and green dots being hidden by red dots. Green dots = patients 2009–2012; blue dots = patients 2013–2016; red dots = patients 2017–2022.

Figure 2

Loading plot of the PCA model of patient-reported outcomes of 7,563 chronic pain patients. EQ5D-index = European Quality of Life instrument; HAD = Hospital Anxiety and Depression scale (A = anxiety subscale; D = depression subscale); MPIS = West Haven-Yale Multidimensional Pain Inventory Swedish version (different subscales); NbPainReg = Number of reported painful regions in the body (0–36); NRS7d = Numerical Rating Scale pain intensity for the last 7 days (0–10); subscales of RAND36 = Short Form Health Survey (SF36) with switch from SF36 to RAND36 in June 2016; TAMPA = Tampa scale for kinesiophobia.

Figure 3

Median (95% confidence interval) of values of the first principal component for each year during 2009–2022 in 7,563 chronic pain patients.

4 Discussion

Medical intuition has been described as a process that is rapid and unconscious, is context-sensitive, comes with practice, involves selective attention to small details, cannot be reduced to cause-and-effect logic, and addresses, integrates, and makes sense of multiple complex pieces of data [15]. Arguably, few clinical practitioners would say that “hunches” based on extensive clinical experience are unimportant or illusory. On the other hand, the possibility of errors of judgment should also be acknowledged. Cognitive errors such as (among others) premature closure, confirmation bias, or the framing effect can cloud the judgement of clinicians [16,17,18]. Eventually, ars medicina (the art of medicine) must be built on evidence and rigorous science – hence, the importance of concepts such as evidence-based medicine and precision/personalized medicine. In the so-called hierarchy of evidence, it must be remembered that expert opinion (including “clinical hunches”) is at the base of the “pyramid”, superseded by case reports/series, followed by observational studies such as the present one, then randomized controlled trials, and finally systematic reviews and meta-analyses as the highest quality evidence [19].

The present study demonstrates that “clinical hunches” should not automatically be viewed as truths. The clinical impression of PRC health care personnel, according to which patients referred to as PRC have gotten more complex and “difficult” over time, could not be confirmed by the data. Importantly, it has to be acknowledged that the hypothesis of PRC patients getting more complex and “difficult” over time was plausible given the now more than a decade old differentiation in Sweden between the IPRP effectuated in specialist pain care and IPRP effectuated in primary care [20]. In 2011, national guidelines were published to ensure the assessment and treatment of chronic pain patients at the appropriate level (specialized pain care or primary care) [21]. So, a question arises whether clinicians at PRC are nowadays seeing a selection of more “difficult” patients, as “easier” patients are now being treated in primary care IPRP? However, appealing as this hypothesis may be, it has (as we have seen) no basis in actual SQRP data. Given the fact that a fifth of the adult population suffers from chronic pain, there is a huge “reservoir” of chronic pain patients that are “referable” from primary care to specialist care, and one can therefore envisage that the implementation of primary care IPRP did not by itself alter the referral pattern to PRC. Be that as it may, is it not conceivable that the clinical “hunch” of PRC personnel might perhaps be a false impression induced by the knowledge that IPRP is now in place in primary care? An alternative explanation could be that increased clinical competency over time has made health care personnel more attuned to the difficulties of individual patients. Consequently, they may perceive patients as “more difficult” when, in reality, it is their capacity to identify problems that has increased.

Importantly, 57 patients were excluded because they were multivariate strong outliers. By definition, we hence know that these 57 patients were “atypical” in terms of their PROMs. However, the proportion of strong outliers did not differ when dichotomizing the study period into early 2009–2015 and late 2016–2022. Thus, a changing proportion of strong outliers over time can hardly explain the “clinical hunch” of PRC health care personnel.

The limitations of the present work should also be acknowledged. First and foremost, one could argue that it is not self-evident that data from SQRP are a valid method to detect what clinicians at PRC feel to be true about their patients. Perhaps the clinicians are right after all. While the lack of a definitive method to assess this issue must be acknowledged, it is also important to ask the question: Given the alleged magnitude of the phenomenon described by PRC clinicians, is it sensible to think that this would not have translated into measurable data in SQRP?

5 Conclusions

Using PROMs available in the SQRP, it is not possible to confirm the clinical impression of PRC health care personnel according to which patients referred to PRC have gotten more complex and “difficult” over time. While it is important to listen to “clinical hunches” emitted by experienced clinicians, it is also essential not to be too quick to equate such impressions with a true state of affairs. Instead, “clinical hunches” can be viewed as hypotheses that should be tested in the context of a research project.