Introduction

Numerous studies suggest that patients’ expectations regarding the likelihood of recovery play an important role in clinical outcome; this evidence pertains to a diverse range of conditions [1] including back pain [2, 3]. The evidence demonstrating the relevance of expectations comes from several sources. Theoretical psychology conceptualises expectations as a type of belief or perception related to future events, the importance of this in a clinical context being the link between expectancy beliefs and (health-related) behaviour [4, 5]. Considerable experimental work has also been conducted, particularly showing the effects of manipulating expectations in exploring the mechanism of placebo effects [6]. There is also a large and growing body of work in clinical populations, a systematic review conducted in 2001 [1] reported on studies showing an association between expectations and outcomes as diverse as analgesic use, return to work, function, exercise tolerance and weight loss.

However, numerous issues relevant to the field of expectations are at present unclear, these include questions regarding definitions, measurement, important related constructs and causal pathways [7]. Not least, it is unknown whether recovery expectations are a stable factor or something that has a potential for changing over time, and if the latter, to what extent any changes reflect the course of the health condition. This question has potentially important implications for clinicians, who might seek to change expectations via their interventions.

Understanding the nature of the relationship between expectations and symptoms offers a number of clinically relevant benefits. It seems that expectations are potentially important in the context of offering individualised prognostic advice to patients since they are shown to be associated with later outcome. Thus, expectations may be sensitive prognostic indicators regardless of whether they have a causal influence on that outcome. If patient expectations do have a causative influence on outcome, then designing interventions to target expectations may be clinically beneficial. On the other hand, if the causal relationship is in the other direction, whereby change in symptom severity is responsible for change in expectations, then there is likely little to be gained by designing interventions to directly target expectations.

In the event that patient expectations do change over time, it might be reasonable to hypothesise that expectations would become more positive as clinical status improves. Yet, it is unclear as to whether this is the case, as relatively few studies have tracked recovery expectations over time along with fluctuations in clinical symptoms [8]. One exception to this is a recently published study by Carstens et al. [9] who found that expectations do not change over time in the majority of people with low back pain. However, they did identify a small group of people for whom a decrease in expectations over time was associated with poorer outcome. This study will further explore the question of whether changes in expectations are associated with changes in pain and disability and if so, whether one precedes the other.

The aim of this study was to investigate the relationship between expectations and symptom severity (pain and disability) in people presenting to primary care with low back pain. The specific research questions were as follows:

  1. 1.

    Do expectations regarding likelihood of recovery change over time in patients seeking care for LBP?

  2. 2.

    Do the course of expectations and the course of symptom severity follow the same pattern over time?

  3. 3.

    Does early change in expectations influence later course of symptoms or vice versa?

Materials and methods

Data from the Nordic Institute of Chiropractic and Clinical Biomechanics chiropractic (n = 926) and GP (n = 241) cohorts were used for this study [10]. Patients in these cohorts self-referred to their primary care practitioner with a complaint of non-specific low back pain. All patients received treatment as usual from their chiropractor or GP, treatment was not constrained as part of the study and generally involved: manual treatment, exercises, and advice for the chiropractic cohort and medication, exercise, advice and referral to physiotherapy for the GP cohort. Study assessment time points were at Baseline (on the day of the 1st consultation), 2 weeks and 3 months. In the chiropractic patients baseline questionnaires were filled out prior to the first consultation, and in the GP cohort, shortly after the first consultation.

The local ethics committee assessed the study protocol and determined that the study did not require formal ethical approval according to Danish guidelines (Danish National Committee on Biomedical Research Ethics. Guidelines about Notification. http://www.cvk.sum.dk/English/guidelinesaboutnotification.aspx.2011/10/04/, 2011).

Participants

Patients aged 18–65 years seeking care for the first time due to the current episode of non-specific LBP, who could read Danish, and completed the expectation question could be included in the cohort. In the chiropractic cohort, additional exclusion criteria were suspicion of inflammatory or pathological pain, nerve root involvement requiring acute referral to surgery, pregnancy and having had more than one health care consultation due to LBP within the previous 3 months.

Measures

At all assessment points participants rated their back pain in the last week on a 0–10 Numerical rating Scale (NRS) [11], back pain-related disability on the 0–23 Roland Morris questionnaire recalculated as a percentage score 0–100 [12, 13] and an outcome expectations measure. Outcome expectations were measured with a single question: “How likely do you think it is that you would be fully recovered in 3 months?” scored on a 0–10 scale where 0 was: ‘Not at all likely’ and 10: ‘Very likely’ [10]. As far as we are aware, the clinimetric properties of this measure have not been comprehensively assessed.

Analyses

All analyses were carried out separately for the chiropractic and GP datasets, this decision was made primarily due to the differences in clinical characteristics between the two cohorts (Table 1). Latent class growth analysis was performed using MPlus 6.1 to categorise participants into groups according to the longitudinal profile of their expectations over the follow-up period, treating expectations score as a continuous measure. Latent class growth analysis assumes that people can be placed into groups with a similar course of expectations over time, the aim is to arrive at the smallest number of profiles that best explains the variability in the complete dataset. The analysis then provides a posterior probability of membership to a profile for each person [14].

Table 1 Participant characteristics
Full size table

Latent class growth models were fitted successively starting with a two-profile model adding another profile at each step. The choice to stop adding profiles was based on the balance of several factors; the model-fit indices, the number of members in the smallest profile, posterior probabilities of assignment to groups and interpretability of the observed profile plots [15]. Latent class growth analysis enables researchers to constrain the variability of the intercept and/or the slope of the best-fit line that defines each cluster. Doing so can improve the fit of the model to the data, the decision whether to do so is based on visual inspection of plots. For these models the variability of the slope was fixed to 0 for each cluster.

Once participants were assigned to their most probable profile, baseline characteristics were compared between the profiles. Differences in age, gender, duration of back pain and number of previous episodes were tested using ANOVA. Pain and disability scores at all assessment points were graphed along with expectation scores for each profile and assessed by visual inspection to assess the longitudinal relationship over time. This qualitative assessment was made in the context of the hypothesis that higher pain/disability scores should correspond to lower expectations and vice versa. Further, over time decreasing pain/disability scores should correspond to increasing expectations (and vice versa).

Two sets of linear regression analyses were performed to investigate whether early changes in expectations influence later change in pain and disability or vice versa. They first addressed the question whether change in pain followed a prior change in expectations independent of a change in pain during that period. This involved construction of a regression model with change in pain from 2 weeks to 3 months as the dependent variable; change in expectations between baseline and 2 weeks as the independent variable and change in pain from baseline to 2 weeks as a covariate. The significance and magnitude of the regression coefficient for the independent variable were inspected to assess the relationship. A separate model was created inputting disability in place of pain.

The second set of models investigated whether change in expectation was associated with a prior change in symptoms and involved change in expectations from 2 weeks to 3 months as the dependent variable; change in pain from baseline to 2 weeks as the independent variable, and change in expectation from baseline to 2 weeks as a covariate. The same model was constructed substituting disability for pain.

Distributions of the variables used in the regression models were inspected for normality. Regression analyses were performed with SPSS20.

Results

Characteristics of the samples

While clinicians involved in enrolling participants in the study were asked to include consecutive cases, no register of the number or reasons for non-inclusion was made. As such we cannot be certain that the included subjects form a representative sample of patients with LBP presenting to primary care. Subjects with missing data on any of the measures were excluded from the analyses. Pain levels in participants were moderate to high on average and disability levels moderate (Table 1). Patients presenting to the GPs were slightly more often female, had more previous episodes of LBP, had longer duration of the current episode and scored slightly worse on the self-perceived general health measure. These features suggest a population with a poorer prognosis than that presenting to the chiropractors. On average patients presenting to the GPs had lower expectations regarding likely recovery than those presenting to chiropractors (Table 1). For these reasons of clinical heterogeneity, the decision was made not to pool the two datasets, but conduct the analyses and the present results separately.

Missing data

Subjects with expectations data missing at any of the time points were excluded from the analyses. This resulted in exclusion of 249 and 44 subjects from the chiropractic and GP cohorts, respectively. In both the cohorts, baseline demographic and clinical data from the excluded participants indicate that these participants were comparable to those who remained in the analyses (data available on request).

Aim 1: Do expectations regarding likelihood of recovery change over time in patients seeking care for LBP? Latent profile analysis

Results from the latent profile analysis suggest a model with seven clusters for the chiropractic cohort and five clusters in the GP group (Table 2). For the seven cluster solution in the chiropractic cohort, 641 of 677 subjects had a posterior probability of >75 % of belonging to their assigned cluster and 19 people were classified in the smallest cluster. For the five cluster solution in the GP cohort, 186 of 197 subjects had a posterior probability of >75 % of belonging to their assigned cluster and the smallest cluster had 17 people. The characterisation of the clusters is descriptive, the labels ‘high’, ‘moderate’ and ‘low’ were not defined a priori and are used to describe the levels of expectations in the clusters relative to each other.

Table 2 Latent profile clusters—Chiropractic cohort
Full size table

The seven clusters in the chiropractic cohort include three that maintain a fairly constant level of expectations, but differ in how positive their expectations are: High (65.3 % of the cohort), Moderate (6.2 %) and Low (5.8 %). Three clusters show a pattern of reducing expectations over time. Two of these clusters start with high expectations at baseline show a small drop from baseline to 2 weeks, and after 2 weeks the High–Moderate cluster (12.4 %) drops a small amount and the High–Low cluster (3.3 %) drops a large amount. The Moderate–Low cluster (2.8 %) begins with medium levels of expectation which drops to low levels by the end of the study period. The final cluster had Increasing expectations over the course of the study (4.1 %). The five clusters in the GP cohort (Table 3) include three with a constant level of expectations, High expectations (33.7 % of the cohort), Moderate (21.8 %) and Low (23.6 %). One cluster showed a pattern of reducing expectations from Moderate–Low (11.7 %). The final cluster showed a pattern of Increasing expectations (9.2 %).

It is noted that >75 % of patients were categorised in clusters that showed very little change in their levels of expectations over the study period. Further, approximately 80 % of the chiropractic cohort reported high levels of expectations at baseline, indicating possible ceiling effects of the expectation measure, with the implication for the regression analyses described below that change in expectations could almost solely occur in one direction.

One-way ANOVA results revealed no significant differences between the clusters in either cohort based on gender. There was a between-cluster difference for age in the chiropractic, but not the GP cohort (Tables 2 and 3). There were significant differences between the clusters based on duration of their symptoms at presentation and number of previous episodes of back pain. Most notably, participants in the high stable cluster (in both cohorts) were least likely to report chronic symptoms and least likely to have experienced more than three previous episodes.

Table 3 Latent profile clusters—GP cohort
Full size table

Aim 2: Do the course of expectations and the course of symptom severity follow the same pattern over time? Visual inspection of plots

To explore the relationship between symptom severity and expectations, the course of pain intensity and disability was tracked longitudinally along with expectations level for the different clusters (Fig. 1). The relationships were assessed qualitatively via visual inspection of the plots. The patterns were very similar for the two cohorts and in all clusters the course of pain and disability was almost identical. The latter being the case, data are only discussed for pain, disability data are available on request from the corresponding author.

Fig. 1
figure 1

Course of expectations, pain and disability

Full size image

In the clusters with stable expectations (High, Moderate, Low) pain and disability reduced gradually over the study period, the High expectations clusters ended with very low symptom severity levels, whereas the Moderate and Low expectations clusters ended with moderate symptom levels. In the clusters with a course of decreasing expectations (High–Moderate, High–Low, Moderate–Low), pain and disability levels also decreased with time, but ended with moderate levels of symptom severity. In the clusters with increasing levels of expectations (Increasing), pain and disability levels reduced over time and ended with low levels in chiropractic cohort and moderate levels in the GP cohort. ANOVA analyses (Table 4) showed that differences in pain scores between the clusters were more pronounced at final follow-up, than differences at baseline or in change in pain.

Table 4 Baseline, follow-up and change in pain scores for the latent profile groups
Full size table

These patterns indicate that changes in expectations and symptoms severity do not always mirror each other closely. However, expectations cluster membership is related to final (3 month) symptom severity, in particular it appears that the large group of patients with consistently high expectations experience the largest improvement in symptoms and enjoy the best final outcome. The patients in the clusters with reducing levels of expectations reported a pattern of large improvement in symptoms in the first 2 weeks, but then little improvement thereafter. The small cluster with increasing expectations reported a pattern of more consistent symptom improvement over the 3 months.

Aim 3: Does early change in expectations influence later course of symptoms or vice versa? Regression analyses

  1. (a)

    Does early change in expectations predict symptom severity?

Regression models were constructed to test the association of early change in expectation with later change in pain and disability (Table 5). The same pattern of results was observed for both cohorts. When the association was adjusted for early change in symptom severity (from baseline to 2 weeks), there was no significant relationship between early change in level of expectations and late change in pain or disability. It appears that early changes in expectations do not independently predict later course of symptoms.

Table 5 Regression analyses: predicting late change in pain and disability
Full size table
  1. (b)

    Does early change in symptom severity predict expectations?

Regression models were constructed to test the association of early change in symptom severity with later change in levels of expectations (Table 6). The same pattern of results was observed for both cohorts. When the association was adjusted for early change in expectations (from baseline to 2 weeks), there was no significant relationship between early change in severity of pain or disability and late change in expectations. It appears that early changes in symptom severity do not independently predict later course of expectations.

Table 6 Regression analyses: predicting late change in expectations
Full size table

Discussion

The most striking finding from this study is that expectations were generally high and remained quite stable over the whole 3-month study period for almost 80 % of participants. Still, increasing as well as decreasing expectations over time were observed in about 20 % of participants. No associations were demonstrated between change in expectations and change in symptoms in the linear regression analyses.

For those with stable expectations one may contend that the levels of expectations were relatively well-aligned with their eventual outcome with regard to symptom level. On average, those with consistently high expectations experienced a large improvement and ended the study period with low symptom levels, those with moderate, stable expectations had some improvement on average but finished with moderate symptoms and those with low, stable expectations experienced little improvement and also finished with moderate symptom levels. Keeping in mind that patients were asked about expectation of full recovery, we can speculate that the level of symptomatic improvement has a reinforcing effect on levels of expectations in these patients.

The situation for those with decrease in their expectation levels over time—approx. 15 % of the total cohort—is somewhat less easily interpreted. Three clusters in the chiropractic group and one in the GP group showed a decrease in levels of expectations over time. In all clusters, the improvement in symptom severity was greater in the first 2 weeks than in the subsequent period (2 weeks to 3 months), this later period of relative symptom stability corresponded to a decrease in expectation levels. It appears that expectations remained high while symptoms were improving, but fell substantially when the rate of improvement decreased. This may reflect dissatisfaction with the clinical course after an initial period of rapid improvement. Regardless of their absolute expectations levels at baseline or follow-up all clusters ended with low to moderate symptom levels (pain score approx. 2–3 on a 0–10 NRS). This explanation would be consistent with the hypothesis that expectation levels are more likely to be a consequence of symptom level, rather than the other way around.

A small group of people in each cohort, around 5 % in total, reported an increase in expectations over time, most noticeably between the 2-week and 3-month follow-up. On average, these people experienced substantial improvement, especially after week 2, and ended with very low symptom levels in the chiropractic cohort, but only small improvements in the GP cohort. It may be that patients were pleasantly surprised when symptoms started to improve and thus gained a more positive outlook.

There were substantial differences between the clusters in terms of their pattern of back pain prior to entry into the study. Members of the clusters with high stable expectations were less likely to present for care with long-term symptoms and were also less likely to have experienced more than three episodes of back pain previously. Clearly, prior experience with back pain is influential in setting initial expectation levels. The question of whether the history or expectations are more strongly associated with outcome in this cohort is addressed in another study [10].

Our results broadly align with those of a recently published study that used cluster analysis to characterise people with LBP according to their expectations. Carstens et al. [9] followed a sample of 281 subjects with occupationally related LBP and found that 85 % had stable levels of expectations over their study period. They also identified a cluster of people with decreasing expectations over time (15 % of their sample), but no increasing cluster. Another longitudinal study [16], also reported that 66 % of their patients with acute orthopaedic trauma had stable expectations, and similarly to our study found a group of patients whose expectations increased.

In attempting to investigate ‘the chicken or the egg’ question, we performed a series of regression analyses to explore whether early changes in expectations predicted later change in symptoms. We found no such association, due at least in part to the fact that there was very little change in expectations during the first 2 weeks. While there were changes in symptoms over the first 2 weeks this did not predict later change in expectations. There might be associations in the individual clusters, hidden in the regression analysis of the whole sample, since the large majority of the sample has stable expectations. This study did not have power to investigate cluster-stratified associations, but it might be of interest for future studies to investigate the subgroups with changes in expectations in more detail.

As mentioned, expectations were stable in most patients and baseline expectation levels did not vary to a great degree between the clusters. This may be taken to suggest that patients are able to quite accurately predict their final outcome, regardless of initial symptom severity. It is possible that individuals are able to access prognostic information relevant to their own health condition that is not captured by other measurement instruments. Alternately it could be that expectation levels are more like a sort of personality trait that remains stable independent of symptomatic fluctuations.

The study has limitations. Importantly, we had only three data collection points, which meant that information about variations over time may have been missed. It is quite possible that greater frequency of measurement is necessary to disentangle the relationships between expectations and symptoms, this is particularly relevant to understanding any potential causal relationships. While there is evidence to support the prognostic value of the expectation measure (under review), there is still uncertainty as to the clinimetric performance of the measure, particularly in terms of reproducibility and responsiveness. It can be questioned whether an instrument that generates such a skewed distribution of observations as observed is an adequate measure. There are also limitations inherent in the conduct of latent class growth analysis. In particular, subjective decisions are necessary in determining the optimal number of clusters; cluster profiles are presented as group means which do not explain all the variability in the sample, and clinical interpretation of the profiles can be debated. Further, interpretation of the relationship between the expectations and symptoms was also a subjective process involving visual inspection of the plots, rather than a formal statistical analysis. Lastly, it is acknowledged that this is a secondary analysis; the protocol for this study was not prepared until after the data were collected.

While there is a growing body of literature that addresses questions relevant to the association of patient expectations with outcome, numerous questions remain unanswered. This study adds to the growing evidence that outcome expectations appear to be a relatively stable construct in a large proportion of patients. While this is not to say that expectations cannot be changed, it does suggest that they are unlikely to do so as part of the normal clinical course for most patients. Of course the question of whether there is any reason to attempt to change expectations depends to an extent on whether they have a causal influence on outcome. We attempted to investigate this by means of a series of regression analyses, we did not find such a relationship, but would urge circumspection in interpretation given the limitations outlined above.

Conclusion

The analysis of two clinical cohorts showed that while levels of expectations vary amongst people presenting for care with low back pain, for the vast majority of people they do not change over the initial 3 months of their condition. For the large group of people with stable and very positive expectations this corresponded to a large improvement in symptoms over time and a good clinical outcome. In general, those with lower, but still stable, levels of expectations experienced more moderate change and less optimal final outcomes. Those whose expectations decreased over time experienced large improvements in the initial period after study entry and little improvement thereafter, the fall in expectation levels appeared to correspond temporally with this period of relative symptom stability. A small group reported increasingly positive expectations over time, the increase in expectations appeared to correspond with the greater symptomatic improvement, in later time period. While these interpretations are suggestive of a causal role of symptom severity for expectations, the question remains elusive. When we attempted to investigate this issue via regression analyses, we were unable to provide compelling evidence in favour of a hypothesis suggesting one causal direction or the other.