Differences across health care systems in outcome and cost-utility of surgical and conservative treatment of chronic low back pain: a study protocol

This study aims to identify differences across health care systems in the Swiss and New Zealand populations such as different compensation schemes, patient characteristics, and co-interventions that modify the effects of surgical and conservative treatment of patients with CLBP. In addition, differences in cost-utility will be identified for different groups of patients.

Three hypotheses will be examined:

1. Differences across health care systems result in better treatment outcomes of surgical and conservative treatment of CLBP in countries with a NHS.

2. Patient characteristics, such as high job satisfaction and short pain duration, and co-interventions, such as cognitive behavioural therapy and return-to-work programs, increase the effectiveness of surgical and conservative treatment for CLBP.

3. Cost-utility in terms of quality-adjusted life years (QALYs) is higher in conservative than in surgical treatment of CLBP.

Previous inconclusive findings may have resulted from potential effect modifiers that varied across different health care systems. For instance, expectations to return quickly to work may function as an effect modifier depending on the health compensation system. This study will examine whether the results pertaining to hypotheses 2 and 3 vary by health care system.

In this multinational prospective cohort study, patients will be recruited consecutively from orthopaedic spine units, rehabilitation centres, and pain clinics in Switzerland (CH) and New Zealand (NZ). Participants will be assessed pre-intervention and at six-month follow-up. The duration of treatment will be a maximum of six months per patient.

The study protocol has been approved by the Ethics Commission of the Canton of Berne (KEK-Gesuchs-Nr.: 191/07).

We defined broad inclusion criteria to ensure that the spectrum of patients represents the spectrum seen in routine settings: CLBP with a duration of the current episode of at least three months, no relevant improvement under prior conservative treatment for one year, patients about to undergo surgical or further conservative treatment, age 18–65 years at beginning of treatment, good understanding of German in the Swiss patient sample or English the New Zealand sample, and written consent (see table 1).

Exclusion criteria are specific CLBP (as infection, tumour, osteoporosis, ankylosing spondylitis, fracture, deformity, inflammatory process, and cauda equina syndrome), acute or subacute LBP (defined as LBP continuing for less than six respectively twelve weeks at beginning of treatment [1, 4], comorbidity that determines overall well-being (e.g. painful/disabling arthritic hip joints), pregnancy, expected loss to follow-up (e.g. due to removal), and unwillingness to complete questionnaires (see table 1).

Surgical interventions for CLBP in this study are fusion (anterior and posterior lumbar interbody fusion, posterolateral instrumented fusion) and nonfusion surgery (decompression, dynamic stabilization, total disc replacement). Given the nature of interventions, a sham intervention is not feasible.

We defined the absence of a surgical treatment as conservative treatment. Different conservative treatments are considered as co-interventions. These co-interventions that also can be applied to the surgical patient group are biomedical, psychologically oriented (cognitive behavioural therapy), and socio-occupational interventions (return-to-work program). Biomedical interventions include physical treatments, exercise therapy, manual therapy, back schools, pharmacological procedures (analgetics, NSAID, opioids, antidepressants), and invasive nonsurgical treatments (acupuncture, injections, facet/nerve root blocks, neuroreflexotherapy, percutaneous electrical nerve stimulation, radiofrequency, intradiscal electrothermal therapy, intradiscal radiofrequency thermocoagulation, cryotherapy of the facets, alcohol denervation, temporary epidural catheters, permanent intrathecal pain-control pumps, and spinal cord stimulators).

The following demographic and baseline characteristics will be collected by patient self-assessment: gender, age, comorbidity (SCQ), marital/relationship status, smoking behaviour, onset/duration/recurrence of LBP, number/level of previous spine surgeries, duration of previous conservative treatment, educational/work/disability compensation status, work history, biomechanical workload, duration of work-absenteeism, job satisfaction, work stress factors, job control, social recognition of LBP from supervisors and colleagues, pain tolerance during work, belief that work has caused LBP, recovery/emotional distress after work, social stress factors, search for social support, condition-specific disability (ODI), general physical and mental health (SF-36), cost-utility of treatment (index based on EQol-5D in combination with cost data), intensity of back and leg pain (VAS), pain frequency, location and number of other painful body sites (SEQ pain assessment), pain behaviour, pain control, self-efficacy, family reinforcement of pain, interference of LBP with work and/or activity, symptom-specific well-being, general well-being, somatisation (MSPQ), depression (ZUNG), fear avoidance beliefs (FABQ), coping strategies (CSQ), and patients' expectations.

Physician-based assessment of patient characteristics at baseline include centre/clinic of treatment, in/outpatient treatment, height, weight, and medication.

Relevant outcomes are identified as scores organised in categories of answers in standardised multi-dimensional assessment tools. Significance of outcomes is defined by effect sizes (i.e. differences between treatment groups in changes from baseline to endpoint, typically divided into three groups: short-, medium-, and long-term changes of respective scores). For this study, outcome measures will be assessed at six-month follow-up. Longer follow-ups may result in difficulties for the identification of effect modifiers [23].

ODI [24, 25], SF-36 [26], EQol-5D [27], and pain measured by VAS [28] are among the most frequently used outcome assessment measurements of CLBP. All of them are valid and reliable. In this study, the primary endpoint at six-month follow-up is ODI. Key secondary endpoints are SF-36, EQol-5D, and VAS.

Other secondary outcomes that will be collected by patient self-assessment are onset/duration/recurrence of LBP [29‐31], work/disability compensation status [32, 33], duration of work-absenteeism [34], location and number of other painful body sites [35, 36], interference of LBP with work and/or activity, symptom-specific well-being, general well-being, and satisfaction with treatment.

Physician-based assessment of secondary treatment outcomes at six-month follow-up include rate/type/therapeutic consequences of complications [37], change in medication, and overall outcome (McNab criteria).

Given the nature of interventions, blinding is not feasible. Outcome assessments are based on standardised, reliable, valid tools.

Possible effect modifiers are all co-interventions, as well as all demographic and baseline characteristics [23, 38‐40], which include gender [41], age [42], severity of pain [43], comorbidity [44], socioeconomic status [45], work-related factors [34, 46, 47], indirect reward parameters due to the health care system [48], psychosocial distress [49‐52], somatisation and depression [53, 54], fear-avoidance beliefs and coping strategies [55, 56], and expectations [57].

The socioeconomic objective of the study is to evaluate the cost-utility outcomes of surgical and conservative treatment of CLBP [8, 9, 27]. The costs considered include the direct health care expenditures borne by the health care system (insurance, public health care) and out-of-pocket expenditures by patients (medication, alternative medicine, personal prevention activities). The indirect social costs due to lost productivity because of absenteeism also will be accounted for [58]. Direct medical costs will collected from health insurance companies in CH, the NHS in NZ and the clinics in which the patients are selected. Indirect costs and out-of-pocket costs will be measured with a cost diary developed by the Zurich University of Applied Sciences in cooperation with the MEM Research Center. The cost diary will be kept by the patients on a weekly basis covering the total period of treatment with a maximum of six months. At beginning of treatment and end of follow-up period utility will be measured with an index based on EQol-5D and SF-36. Changes in the index will be translated into changes in QALYs.

The cost study of CLBP patients will concentrate on the segment of high-cost patients and address this question: Is this group homogeneous with respect to their socioeconomic status, care-seeking and back pain patterns? If the answer is no, are there subgroups whose cost may be reduced substantially by appropriate (nonmedical) support?

The cost-utility part of the study will be done in collaboration with the Zurich University of Applied Sciences, Institute of Health Economics.

The backbone of this Swiss-based study is the IT-infrastructure of the MEM Research Center, University of Berne, which has developed and improved documentation systems and questionnaires over decades.

Orthopaedic spinal units, rehabilitation centres, and pain clinics in CH and NZ will participate in this study. Collaborators in the main clinics will act as local coordinators in association with local research nurses. The outcome assessment is based on physician- and patient-derived data.

After written informed consent patients will be examined by the surgeons or physicians under whose care they are. Research nurses will hand out self-assessment questionnaires to the patients and will be available to help completing these questionnaires if necessary. The completed patient-based questionnaires will be collected by the research nurses and, together with physician-based questionnaires, forwarded to the MEM Research Center in CH respectively to the University of Otago in NZ. All data will be saved at the central database of the main server at the University of Berne.

The computerised medical data collected at the MEM Research Center do not contain names. Only the patient number will be able to connect the names and the medical information at the treatment level. The servers have firewalls and data encryption, and are located in a physically secure environment.

In this prospective cohort study, two interventions for CLBP are compared: surgical and conservative treatment. Subgroups of different surgical treatments can be distinguished. Nonsurgical co-interventions may have an effect on patient groups with surgical as well as with conservative treatment.

Follow-up data at six months will be analysed according to the intention-to-treat principle, except for those patients who change treatment groups before the beginning of the allocated treatment.

Changes in dependent outcome variables between baseline and follow-up at six months will be analysed by multiple linear regression models. Surgical and conservative treatment, patient characteristics, and co-interventions are independent variables. To address potential effect modification, we will then incorporate interactions between type of treatment and the different co-interventions and patient characteristics.

Highest priority among co-interventions will be given to the analysis of effects of cognitive behavioural therapy and return-to-work programs. Among patient characteristics highest priority in data analysis will be given to gender, age, and duration of work-absenteeism.

The most important cost predictors (including cost drivers) for total, direct, and indirect costs are identified by multivariate modelling approaches.

One multivariate modelling approach is to determine which individuals with CLBP are in the top, say, 15% of the total cost, and to contrast this group with the remaining individuals. The relevant cost predictors are identified by discriminant analysis, or any other suitable classification method such as logistic classification or classification trees.

Another approach is to model the costs by a multiple regression type model, where the costs are assumed to be log-Gaussian or gamma distributed. The uncertainty in the prediction of cost will be quantified by state-of-the-art methods [59, 60].