Feasibility and long-term efficacy of a proactive health program in the treatment of chronic back pain: a randomized controlled trial

The study was conducted as a parallel group randomized controlled trial using Zelen’s design. The specific characteristic of Zelen’s design (also called randomized consent design) is that consent to participate is sought only after randomization [18, 19]. The study adheres to CONSORT guidelines.

Eligible participants were members of the German Private Health Insurance Central with a minimum age of 18 years and showing symptoms and “administrative signs” of chronic BP. They were selected by the employees of the health insurance company on the basis of predefined selection criteria (see Table 1) and analysis of existing billing data on the insured. The selection criteria were chosen in such a way that the identified persons were highly likely to suffer from chronic BP. The billing data used for this included treatment and cost information on outpatient treatment (e.g. drugs), inpatient treatment (e.g. surgeries) and daily sickness allowance. Conclusions about the disease were drawn from invoices submitted by the insured persons which also included the ICD codes based on which treatment choices had been made [20, 21].

Between April and October 2015, eligible members of the private health insurance were randomly allocated to the intervention (IG) or the control group (CG) by the study center at the university. Simple block randomization was conducted by an independent external researcher using BiAS for windows version 11.02. The allocation ratio was 4 to 3 to compensate for anticipated different participation rates in IG and CG. The private medical insurance invited the allocated members in writing to participate in the study arm to which they were assigned without disclosure of the “pre” randomization step (Zelen’s design). The invitation letter described the target group as persons with BP over several months. Informed consent was obtained from the IG for participation in the health program and follow-up measurements to evaluate the effects, and from the CG, for participation in a follow-up study to evaluate the effects of usual care for chronic BP. Members of IG and CG filled in identical online-questionnaires at home at baseline as well as one and 2 years thereafter. Between April and October 2017 data collection ended with the two-year follow-up.

The main elements of the health program “initiative.back” under evaluation were as follows:

(1) IG-members were advised to consult a physician from a network of back experts (composed of general practitioners, orthopedic specialists, pain therapists, psychotherapists, physiotherapists), all of them following the recommendations of the National Disease Management Guideline on non-specific low BP [8, 9], such as interdisciplinary assessment and multimodal treatment for patients with chronic or recurrent BP. The initial examination by the physician also included investigation of the back. Based on the examination results, a tailor-made therapy program for the back muscles, safe from a medical point of view, was put together for each participant in specialized back centers. Participants received equipment-based training for a maximum of 24 h over a period of three to 4 months. Each of these lasted 60 min and included a combination of strength training, gymnastics and relaxation exercises to strengthen the back muscles and relieve the strain on the spine (FPZ-therapy [22], for details see http://​www.​fpz.​de).

(2) Each IG member received personal health coaching over the phone from an external professional coach (not employed at the private medical insurance). Participants were coached during the treatment phase as well as up to 6 months thereafter in the context of after-care. A maximum of 222 min spread over 16 contacts with each participant was planned, but frequency and duration of coaching over the phone were geared to individual needs. Coaching aimed at encouraging life style changes and the consolidation of physical activities. During aftercare the participants were eligible to receive twice an activity bonus of 100 Euros each if they participated in any sports activities of their choice.

The maximum duration of the total health program was 12 months.

For evaluation of acceptance, the health insurance company provided information on participation in the health program (entire program completed, participation in the program prematurely terminated or program not joined) and on the intensity of use (number of therapeutic exercise sessions, duration of coaching over the telephone).

The CG members did not undergo any study intervention, receiving only “usual care” i.e. care according to the prescriptions of their health care providers (family doctors or medical specialists). Information on care procedures for their BP was not available. Therefore, it is not clear to what extent treatment of BP was in accordance with the recommendations of the National Clinical Practice Guideline for Non-Specific Low BP [8, 9].

Severity of BP as one of the two primary outcomes was assessed by the German version of the Chronic Pain Grade Questionnaire (CPGQ) [23, 24]. The CPGQ is a brief and simple instrument to hierarchically grade the severity of chronic pain in terms of pain intensity and disability and can be used in general population-based studies as well as in those relating to pain patients in primary care. In the presented study we measured intensity of BP and BP-related disability using the recommended scoring rules [23]. Intensity was calculated as the average of three 0 to 10 ratings on current BP, worst BP and average BP (in the past 6 months) and was expressed as a percentage value of 0 to 100% (with higher scores indicating more severe pain). BP-related disability was expressed as disability points. These were determined on the basis of the number of self-reported disability days in the past 6 months (≤6 days = 0 points, 7–14 days = 1 point, 15–30 days = 2 points, ≥31 days = 3 points) and the average of three 0 to 10 ratings on experienced impairments in daily, family/social and work/household activities, expressed as a percentage value of 0 to 100% (≤ 29% = 0 points, 30–49% = 1 point, 50–69% = 2 points, ≥70% = 3 points). Disability points are the sum of points for disability days and impairments in activities and range from 0 to 6 points with higher scores indicating more severe disability. BP severity can be graded in 4 hierarchical classes: Grade I (disability points < 3, pain intensity < 50%), Grade II (disability points < 3, pain intensity > 50%), Grade III (disability points = 3–4) and Grade IV (disability points = 5–6).

Health-related quality of life (HRQoL), the other primary outcome, was assessed with the German Short Form 12 (SF-12) [25], a generic health status instrument. Physical and mental health composite scores were computed, each ranging from 0 to 100, where zero indicates the lowest health status measured and 100 the highest.

Secondary outcomes included the risk of BP chronification measured by the Keele STarT Back Screening Tool, German version (STarT-G). The STarT-G consists of nine items. The first four items relate to biomedical factors and the remaining five identify psychosocial risk factors. A total score (ranging from 0 to 9 points) and a psychosocial sub-score (ranging from 0 to 5 points) are calculated. Patients can then be allocated to one of three prognostic groups using established scoring cut-offs (low-risk: total score ≤ 3 points; medium-risk: total score > 3 and sub-score < 4 points; high-risk: total score > 3 and sub-score ≥ 4 points) [26‐28].

Psychological distress was assessed with the Patient Health Questionnaire-4 (PHQ-4), a 4 item inventory rated on a 4 point Likert-type scale. It is composed of the first two items of the Generalized Anxiety Disorder–7 scale (GAD–7) and the Patient Health Questionnaire-8 (PHQ-8). PHQ-4 total score is determined by adding together the scores of each of the four items, ranges from 0 to 12, with higher scores indicating more emotional distress (anxiety and depression) [29, 30].

Physical activity was measured with two questions referring to the last 3 months: “On how many days are you physically active on average in a way that you start to sweat or get out of breath?” Active participants were further asked: “How long are you physically active on average on these days?” Possible answers were: “less than 10”, “10 to less than 30”, “30 to less than 60” or with “more than 60” min [31].. As outcome parameter we used the number of days per week with at least 10 min of physical activity a day.

The sample size was calculated on the ability to detect a statistically significant difference in the primary outcomes between IG and CG at the 2-year follow-up with a small effect size of Cohen’s d = 0.3, a 2-sided α = 0.05 and a test power of 1-β = 0.8. Anticipating a dropout rate of up to 40%, we aimed at having 290 participants per study arm to ensure a sample size of at least 176 participants per study arm with data at the 2-year follow-up.

Statistical analysis was performed on an intention-to-treat basis. Each participant was analyzed in the study group to which he or she was randomized. Only participants with complete data (baseline and 2-year follow-up) were analyzed. Dropout analyses were conducted to estimate attrition bias. If a question was left unanswered, the participant could not proceed further till it was filled in. The online questionnaire, thus structured, prevents single missing values.

All significance tests were performed without α adjustment. Due to multiple comparisons the results have a descriptive character [32]. Effects sizes for the between-group differences were calculated as Cohen’s d (or Hedges’ g) with 95% confidence intervals [33].

In addition to the primary analyses, subgroup analyses were done to explore the heterogeneity of treatment effects in participants with different risks of BP chronification. For the primary outcomes, we contrasted treatment effects in persons with medium or high risk of BP chronification (STarT-G total score > 3) at baseline with treatment effects in persons with low risk (STarT-G total score < 3). The between-subgroup interaction test of Altman was used to assess if potential treatment differences depended on the person’s subgroup [34‐37].

Statistical analyses were performed using IBM SPSS Statistics 22. For the computation of effect sizes, the free software “Psychometrica” was used [38].

Written informed consent was obtained from all study participants. The independent research ethics committee of the University of Lübeck gave approval for the study (Re.-No.14–249, dated 20 Nov 2014). The procedure for collecting and processing the study data was agreed upon with the data protection officer of the private health insurance company. The contract research study was supervised by the Lübeck research group within the framework of the contract with the insurance company. The trial was registered at the German Clinical Trials Register under DRKS00015463 retrospectively (dated 4 Sept 2018).

A total of 3462 insured persons were randomized and contacted. Of these, 552 gave their consent to participate in this study. The participation rate was significantly lower in the IG (N = 258, 13.1%) than in the CG (N = 294, 19.6%) (p < 0.001). The follow-up questionnaire was completed by 444 (80.4%) participants 2 years later. The IG and CG showed different dropout rates (IG: 26.7%, CG: 13.3%, p < 0.001) (see Fig. 1).

Table 2 shows participant characteristics at baseline.

IG and CG members showed comparable sociodemographic characteristics. Significant differences were seen in severity of BP (IG worse than CG), in the risk of BP chronification (IG higher risk than CG) as well as in satisfaction with medical care of BP (IG less satisfied than CG).

Analyses were done for study participants. Because of the different drop-out rates in IG and CG, the study groups were analyzed separately. There were few significant differences in the demographic and clinical characteristics at baseline between responders and those lost to the 24-month follow-up (non-responders). At the baseline, the non-responders in IG as well as CG differed in one of the 12 characteristics listed in Table 2. The non-responders in the IG were significantly more dissatisfied with the previous BP treatment than the responders (4.8 versus 5.7; p = 0.038). There were significantly more men among the non-responders than among the responders in the CG (76.9% compared to 59.6%; p = 0.038).

Among the responders in the IG, the proportion of study participants who completed the health program was significantly higher than among the non-responders (73.5% versus 40.6%; p < 0.001) (see Fig. 1).

Approximately one in eight of the insured persons who were invited to participate in the initiative.back accepted this offer (258 out of 1963). Among these, about 2 out of 3 (167 out of 258) completed the health program, about 7% (17 out of 258) terminated it prematurely, and 28% (72 out of 258) quit the program even before starting on it (see Fig. 1). The most frequently cited reason was the inconvenient distance from the place of residence to the nearest medical practice or training center.

Of those who participated in both the program and the 24-month follow-up, 91% underwent the maximum of 24 h of exercise therapy spread over the entire duration of therapy, 9% received only 10 h. On average, 191 min of coaching over the telephone per capita was realized (SD = 62; range 51–443 min).

As far as changes over time are concerned (see Table 3), in the IG, significant improvements were observed in 6 of the 7 outcomes (excluding mental health) and in the CG, in 3 outcomes (pain intensity, disability and mental health status). All observed positive changes were in the small range (SMR < 0.5).

To assess the long-term treatment effects, we compared the outcome variables between IG and CG at the 2-year follow-up adjusted for baseline differences. In 5 of 7 outcomes, the IG reached significantly more favorable scores than the CG (see Table 4).

In comparison to the CG, the participants of the IG presented themselves at the 2-year follow-up with less BP-dependent disability and demonstrated improved scores in their physical health status (SF-12). There were no significant differences between the groups at the 2-year follow-up in intensity of BP and mental health.

Both the psychological distress (total score of the PHQ-4) and the risk of BP chronification (total score of the STarT-G) were lower in the IG than in the CG. The IG reported more days per week with at least 10 min of physical activity than the CG.

All observed significant differences in the patient-reported outcomes between IG and CG correspond to small effect sizes (range of d: 0.21–0.26).

In addition to the main analysis, treatment effects in the primary outcomes were separately analyzed in two subgroups consisting of study participants with either low risk of BP chronification (STarT-G total score not exceeding 3) or with medium or high risk (STarT-G total score greater than 3) at baseline.

Significant long-term effects only occurred in the subgroup with medium or high risk of BP chronification (Table 5). In this subgroup, intensity of BP and disability (GCPS) were lower and the physical health status (SF-12) was higher in the IG than in CG with effect sizes of approximately 0.4. Only the difference in mental health status did not reach significance.

Altman’s between-subgroup interaction test was used to examine whether this heterogeneity in treatment effects depends on the person’s risk-level of BP chronification at baseline (see Table 6).

The results of the interaction tests suggest that persons scoring higher in STarT Back Screening Tool at baseline benefit significantly more from the health program than persons with low risk scores.

Although health care policy requires scientifically sound evaluation of health care innovations, unproven innovations are too often implemented in health care systems. Since 2016, German statutory health insurances can apply for funds by the newly created Innovation Fund (worth € 300 million per year) for health services-related research projects. However, private health insurances have no access to this fund. It is to the credit of Central as a private health insurance company that they made an effort to get their new health program evaluated and its efficacy examined not in the short term - where effects are generally larger - but in the long term.

Since a conventional RCT design (randomization after informed consent) carries with it a risk of dissatisfaction on the part of the members of the non-preferred arm, a “post randomization consent design” according to Zelen was chosen, which, however, is not uncontroversial [49‐51]. Different participation rates in IG and CG and numerous baseline differences between IG and CG are regarded as typical disadvantages of using such a design. Both occurred in our evaluation study reducing the comparability of the study arms. The invitation of the health insurance company to participate in the health program with accompanying evaluation (IG) was accepted by chance by fewer insured persons than the invitation to participate in a long-term observation of their BP problems (CG). The difference in the willingness to participate in the study is probably due to the significantly different time and personal commitment required from study subjects. Participation in the CG was limited to filling out an online questionnaire several times, while participation in the IG was associated with a variety of requirements (including visits to the doctor, muscle training, telephone calls from the coach).

As is frequently the case in health services research, our study participants could not be blinded to the treatment they received. The only thing they were not told was about the randomized group allocation based on the Zelen’s design we used in our study. The physicians administering the interventions to the IG and those taking care of the CG were not aware of the evaluation study.

Furthermore, only patient-reported variables were used as study outcomes. However, taking into account the absence of any dependency of the participants on the researcher handling the data, the risk of social desirability bias can be assumed to be low.

The influence of possible moderators and mediators such as comorbidity or operations on the outcomes could not be evaluated because such data were not available.

The interesting question of whether sociodemographic variables (such as age, gender, formal education) were (or were not) associated with treatment outcomes remains unanswered, being outside the scope of the study.

A sample of members of a single private health insurance does not provide a representative picture of the German population. As is known [52, 53], members of German private health insurances (about 15% of the German population) differ in sociodemographic and health-related characteristics from members covered by statutory health insurances. For instance, they have better than average levels of education. The study results are, therefore, not generalizable.

An attrition bias cannot be excluded. We considered only complete cases and it might be that persons lost to follow-up at 24 months had better or worse outcomes resulting in an under- or overestimation of true effects. At the 12-month follow-up the drop-out rate was high. However, with reorganization of follow-up management, it was possible to reduce the lost-to-follow-up rates and thus overcome the threshold of 30% set for judgment of risk of attrition bias (see [54]). Hence, any attrition bias in this study was likely not substantial.

Additionally, the possibility that analyses of multiple primary and secondary outcomes could have increased the risk of significant effects by chance (i.e. inflation of α-error) cannot be excluded.

We identified two possible points of improvement for the future use of the health program. On the one hand, before inviting patients to participate in the program, it is necessary to ascertain the extent to which network doctors and associated therapy centers can be found within easy reach of the insured person’s place of residence. Ensuring easy access in terms of distance might increase program acceptance and adherence. Furthermore, the health program offered should not be based on an “one-size fits all” concept. The positive effects might be increased by the use of the STarT back tool to stratify eligible participants with BP into low, medium and high risk of BP chronification with special care pathways for the three subgroups. The predictive and discriminative ability of the STarT back tool in populations with BP of variable episode duration is widely supported in the literature (inter alia [55‐58]). Sophisticated treatment systematically targeting medium and high-risk groups apparently leads to improved outcomes [59]. Our results suggest that the low-risk subgroup derives hardly any benefits from the health program in the long-term; with screening, potential overtreatment of the low-risk subgroup probably needing only minimal treatment can be avoided.

In summary, the available results of the present study support continuing the program. Approaches for increasing the observed beneficial effects have been mentioned above. An analysis of the cost data is pending, so that a final cost-benefit assessment has not yet been carried out.