Does evidence support physiotherapy management of adult female chronic pelvic pain? A systematic review

1.1 Objectives

To examine current evidence for an effect of physiotherapy as a sole intervention or significant component of a multidisciplinary intervention on pain, physical activity and quality of life in adult women with chronic pelvic pain.

2.1 Information sources

The search strategy and index terms were planned by the first author (SL) under guidance of a clinical expert (JN) and the supervisor group (PJ, TT). The search strategy was inspired by the search words developed by the Cochrane specialised register group; i.e. the Menstrual Disorders and Subfertility Group, The Incontinence Group, The Functional Disorder Group and The Musculoskeletal Group.

The primary search was conducted in April 2010, and updated in September 2010, April 2011, and latest in September 2011. This revealed no new relevant studies. The electronic databases Medline (1966–2011), Embase (1980–2011), Cinahl (1982–2011), PsycINFO (1995–2011), and the following evidence-based practice resources; CENTRAL (1987–2011), Physiotherapy Evidence Database (PEDro, Centre for Evidence Based Physiotherapy 2010/2011), and Database of Abstracts of Reviews of Effects (DARE 2011) were searched. The search contained keywords (Mesh and Thesaurus) and was a combination of searches in Keywords and Title/Abstract to ensure that all new studies were included. Multiple keywords to describe CPP and physiotherapy modalities were used reflecting the lack of a consensus definition and of validated diagnostic markers in categorizing CPP [6] and the range of physiotherapy interventions and the terminology to describe such interventions (Table 1).

2.2 Study selection

Titles were scanned by the first author (SL) and obviously irrelevant studies were removed. Abstracts of potentially eligible studies were reviewed by two authors (SL and TT). The first author (SL) is knowledgeable in the area under review, the second author (TT) is not a content expert but has experience with conducting systematic reviews. Disagreements concerning the relevance of the studies for the review were resolved by discussion. A third assessor was available as an arbiter in case of disagreement. Papers were included in this systematic review if they met the inclusion criteria listed in Table 2. The full text versions of all papers that met the inclusion criteria were retrieved for data extraction and risk of bias assessment.

2.3 Critical assessment of included studies

Critical appraisal of each included study was conducted by determining:

• The level of evidence for an intervention effect in the included studies using The Cochrane Collaboration’s tool for assessing risk of bias. The tool is recommended for experimental and controlled studies and involves consideration of six features: sequence generation, allocation sequence concealment, blinding, incomplete outcome data, selective outcome reporting and “other” potential sources of bias. Items in the risk of bias assessment were judged “adequate” (+), “unclear” (?), or having the “potential for bias” (−) for each study (Fig. 2). Blinding of participants in a physiotherapeutic intervention is nearly impossible, and complete blinding of personnel/therapists was considered difficult to uphold. We therefore judged blinding as adequate if outcome assessors were blinded.

  The Cochrane Collaboration’s tool for assessing risk of bias was not developed for non-randomised clinical studies (NRCTs). The six domains included in the tool could usefully be assessed for prospective cohort studies as recommended by the Cochrane Collaboration, but are not necessarily sufficient for assessing risk of bias. As recommended by the Non-Randomised Studies Methods Group (NRSMG) of the Cochrane Collaboration we chose to additionally assess the potential risk of confounding in the included studies that were not RCTs under guidance by NRSMG. We prespecified the following potential confounding factors; age, duration of pain, number of pain sites, depression, sexual dysfunction, earlier pelvic operation for the pain, and history of sexual or physical abuse.

• A categorization of systematic error (bias) of the included clinical intervention studies into levels of evidence proposed by The Cochrane Hepato-Biliary Group, Copenhagen Trial Unit, Rigshospitalet, Denmark. Trials with one or more bias components assessed as inadequate or unclear were considered to be at high risk of bias, while trials with all quality components assessed as adequate were considered to be at low risk of bias [25].

2.4 Data extraction and management

A standard review checklist based on guidelines in the Cochrane Handbook [26] was used independently by the review authors to extract data for each included trial. Information was collected in PICO (S) structure: patients, intervention, control, outcome measure (and time of follow-up) and study design [25]. The following data were reported in “Characteristics of the included studies” (Table 3).

• Study design and country site.

• Type of chronic pelvic pain (unspecified, related to the reproductive system, the bladder or the bowel) and duration.

• Number and characteristics of study participants.

• Description of experimental interventions, including extent and duration; description of control intervention if any.

• Outcomes and time points collected and reported.

• Drop-out rate.

Data were initially extracted from each trial by SL and subsequently verified for consistency and accuracy by TT. We resolved disagreement by discussion until consensus was reached.

Risk of bias assessment of the included studies is summarised in Fig. 2 and Categorization of bias into levels of evidence is described in Table 4.

We used Review Manager (RevMan) version 5.0 (The Nordic Cochrane Centre, The Cochrane Collaboration, Copenhagen, Denmark) for data analysis. Data on the primary outcome (pain) are presented in Tables 5a–c; data on secondary outcomes (physical activity, quality of life) are narratively summarised. Most of the included RCTs compared outcome improvement as within-group pre- and post-treatment results instead of calculating between group differences. We judged between-group analyses to be more informative based on the randomised clinical study design. We calculated mean differences (MD) and 95% confidence intervals (CI) for continuous measurements from comparative studies (Table 5a). For binary (or dichotomous) outcomes, we calculated risk ratios (RRs), the corresponding 95% CIs using Mantel–Haenszels methods (Table 5b), and for significant results the numbers needed to treat (NNT). For non-comparative studies we expressed results for each study as mean changes between pre- and post-treatment with 95% CIs and p-values 3 (Table 5c).

2.5 Types of outcome measures

Core outcomes were based on the recommendations of Cochrane Pain, Palliative and Supportive Care Review Group (PaPaS). We considered measures on pain reduction, quality of life and physical functioning/activity to be important outcomes in this systematic review.

The primary outcome was pain reduction measured on validated pain scores, scales or questionnaires or by descriptive endpoints. Considerable effort has been devoted into quantifying the magnitude of change in pain intensity that is considered clinically relevant to patients with chronic pain [27,28,29]. A moderately important benefit is defined as at least 30% reduction in pain and a substantial important benefit as at least 50% reduction [30]. Clinical studies reporting average changes in pain scores are of limited utility and often inappropriate. The average results represent only a small minority of patients as responses in chronic pain trials frequently take the form of a skewed or bi-modal distribution, where some patients obtain very good pain relief while others obtain very little [31,32,33]. Therefore, when sufficient raw data (individual patient data) were provided in the articles [10,34] we dichotomised improvement in pain scores according to the above clinically relevant criteria.

Secondary outcomes were measures of quality of life (QoL) and physical functioning/activity measured by validated questionnaires or rating scales. We summarised analyses on statistically significant results.

We reported other relevant outcomes such as depression, general pain, associated symptoms, medicine intake and pelvic floor muscle strength as in the original studies in Table 3 (Characteristics of the included studies). Due to substantial heterogeneity in outcome reporting between studies, these outcomes were not pooled.

We considered analyses of reports of sexuality and patient global impression of intervention important for women with CPP and therefore summarised these outcomes as reported by included studies.

Adverse events: We summarised adverse events as reported by included studies.

2.6 Duration of intervention effect and follow-up period

Intervention effects decrease over a period of 2–12 weeks, especially of minimally effective interventions [31]. We therefore considered 10–12 weeks duration of intervention and follow-up as a minimum requirement, not least in light of the chronicity of the condition and the likelihood of a 12-week measurement becoming standard [31].

3.1 Search strategy yield

The search strategy identified 172 citations in MEDLINE, 73 in Cinahl, 1656 in EMBASE, 1542 in Cochrane and 26 in Pedro, yielding 3469 unique citations. Ninety-five out of 140 potentially eligible citations were identified after reviewing the titles of the 3469 citations and after removing duplicates (45 citations). Six additional, potentially eligible studies were found from reference lists in the studies retrieved through the search, nine studies were retrieved by hand search, and two studies were obtained through personal communication. The resulting 112 records were screened and 86 records were excluded on the basis of information provided in the abstracts. Of the 26 remaining, potentially eligible studies, 15 were excluded in accordance with the inclusion criteria (Fig. 1). Eleven clinical trials involving initial recruitment of 782 participants were included (Table 3). One study had different outcomes reported in additional papers; Haugstad reported 1-year follow-up in 2008. We reported outcomes using the main study identifier [10].

Fig. 1

Retrieval and review process.

Seven authors were contacted [35,36,37,38,39,40,41] for information on publication of abstracts, gender- or diagnosis-specific data (raw data), or study design. This lead to preliminary inclusion of two conference proceedings [39], one abstract [38] and one article [37] retrieved through the electronic search strategy. The remaining authors either did not respond or could not give the requested information.

3.3 Critical appraisal

3.3.1 Risk of bias (internal validity)

3.3.1.1 Randomised clinical trials

Three RCTs had level 1b evidence with low risk of bias and three RCTs had level 1d evidence with high risk of bias (Table 4).

Sequence generation and allocation concealment was graded adequate in the publications [10,42,43] or following confirmation from the authors that a robust method of randomisation had been employed [37,45]. Following discussion the review authors judged allocation concealment as adequate in the study by Haugstad [10] despite an unsure random component, i.e. drawing of lots with the patients’ name. Heyman [44] used stratified block randomisation in blocks of four, which may have enabled prediction of allocation (unclear risk of bias).

Outcome assessment and blinding were graded adequate in all RCTs except one where no intention-to-treat analysis was conducted [43]. Outcomes as pre-specified in the article were reported in all the six RCTs except for one [10]; Haugstad included a non-prespecified questionnaire assessing psychological distress and general well-being (GHQ-30) at 1 year follow-up [34]. These outcomes are irrelevant to the outcomes analysed in this systematic review. Hawk reported all outcomes at baseline and at the 6-week follow-up, but failed to report outcomes at the 12 and 24 week follow-up as otherwise prespecified in the article. Three of the six randomised clinical trials were reports of feasibility/pilot studies [37,43,44]. One multicentre study reported substantial deviation from study protocols between the participating centres. These deviations prevented pooled analysis of data from all centres [43]. A crossover RCT [42] did not state a wash-out period between treatment periods, which may have introduced a “carry-over” of treatment effect. This led to a judgement of high risk of bias and hereby a lower level of evidence.

3.3.1.2 Non-randomised clinical trials

All NRCTs were judged at high risk of bias except for one [46], which we judged had a moderate quality level.

Due to the NRCT design, generation of allocation sequence and allocation concealment was irrelevant [46,47,48,49]. Neither participants, personnel nor assessors were blinded in these studies so we judged outcome measures at potential risk of performance and detection bias. All but one NRCT [47] had substantial drop-out at one-year follow-up. The cohort study had 62% non-responders, however a drop-out analysis revealed no difference between responders and non-responders in baseline characteristics [46]. Three NRCTs reported all outcomes as prespecified [46,47,48]. Albert [49] omitted reports of time spent sitting, standing or lying and state of mood, and neither outcome measurements on pain were reported at end of treatment and at 3 months follow-up as otherwise prespecified. Instead, only outcomes at 1 year follow-up were reported.

3.3.1.3 Summary

Fig. 2 illustrates a general high risk of bias across the included studies (one or more component rated as inadequate or unclear). Three randomised trials were rated as level 1b evidence with low risk of bias, three RCTs as level 1d evidence with high risk of bias. The cohort study was rated level 2b evidence and the remaining three non-randomised trials level 4 evidence (Table 4). Studies high on the hierarchy potentially contain less bias than those lower on the hierarchy.

Fig. 2

Risk of bias summary: review authors’ judgements about each risk of bias item for each included study.

3.4 Effect of physiotherapy as a sole or significant component of a multidisciplinary intervention

3.5 Synthesis of results

Neither data gained from RCTs nor NRTCs could be pooled because of heterogeneity of participants, interventions, outcome measures, and follow-up times. This prevented synthesis of results across studies and hereby also meta-analyses.

4.1 Summary of evidence

The results of this systematic review indicate that there is level 1b evidence (low risk of bias) that Mensendieck somatocognitive therapy combined with standard gynaecological care improves pain experience in female CPP [10,34]. The results are clinically relevant with point estimate reductions in pain exceeding 50%. This effect persisted nine months post-treatment. A growing body of evidence is establishing that a 50% reduction in pain experience or more is associated with major improvements in function, sleep, fatigue, depression, quality of life and ability to work [50,52,53]. This supports the significant results on physical function, gait, respiration and sitting posture following somatocognitive intervention [10,34].

There is level 1b evidence (low risk of bias) for an effect of a multidisciplinary intervention including physiotherapy compared to standard diagnostic laparoscopy on the general experience of pain and level of daily activity in women with CPP. However, the minimum level of pain improvement required to demonstrate a statistically significant effect was not specified in the article, and the clinical relevance is therefore uncertain. Notably, no improvement was demonstrated in specific pain measurements (McGill) [45], a finding that is similar to the findings of the comparable intervention study by Lamvu 2006 [46]. Both studies had sufficient follow-up periods (1 year) whereas duration of treatment was only specified in one (6 months).

There is level 1d evidence (high risk of bias) that physiotherapeutic distension of painful pelvic structures combined with pain counselling improves pain experience compared to TAU (1 counselling session). The clinical relevance of this result is however difficult to assess because the experience of pain is evaluated as a group average and because of the short follow-up period.

Insufficient level of evidence is provided to draw conclusions in regard to the effect of psychosomatic group treatment [49], and to Modified Thiele Massage [48], even though studies of these interventions demonstrated a statistically significant pre-post reduction in pain on VAS.

It was not considered possible to meta-analyse the results of the included studies, which is a limitation of this review. Moreover, the majority of the studies included, investigated the effect of physiotherapy in combination with medical or psychological treatment. Therefore the stand-alone value of a physiotherapeutic treatment modality cannot be determined. By reference to the risk of bias assessment of the individual studies (Fig. 2), the categorization into levels of evidence (Table 4) and the clinical relevance of pain improvement, careful consideration must be kept in mind in the analyses of results and the conclusions drawn from the studies.

4.2 Internal validity

Internal validity was assessed in the risk of bias assessment of the included studies. Internal validity was generally limited by small and potentially selected samples, heterogeneous or unspecified patient characteristics (confounding factors, especially in NRCTs), questionable allocation concealment, lack of blinding, inadequate outcome data addressed and selective reporting.

Small sample sizes and substantial variation in measurements of effects (SD) and wide confidence intervals introduce a risk of random error, especially the type II error (β). The consequence is a risk of false acceptance of “no-effect of intervention”. This has to be kept in mind in the interpretation of studies with small sample sizes in which statistical significance (α) is only nearly reached, resulting in rejection of an “intervention effect” [37,42].

4.3 External validity

An overall problem was the lack of flow-charts stating how many women were initially assessed for eligibility and how many of these accepted to participate: FitzGerald (2009) had an unsatisfying presentation of consenting participants (38%), Hawk reported that 92% of eligible women accepted participation, but could not subsequently provide any useful results. The remaining eight papers presented no flow-chart preventing assessment of the potential for an overall risk of selection bias. This poses a potential risk to the external validity of this systematic review and hereby its applicability to clinical practice [54].

Inadequate duration of pain treatment and heterogeneous follow-up times in the included studies also constitute a potential limitation of this review. Chronic pain conditions require treatment for considerable periods of time. In this systematic review only four studies [34,45,46,49] reported follow-up up to 1-year post-treatment, and of these only two RCTs were at low risk of bias. This may affect the external validity of the evidence for long-term treatment effects and hereby implications for practice.

4.4 Implication for practice

The strength of recommendations for clinical practice depends on the level of evidence as indicated by a risk of bias assessment (internal validity), on consistency of results between studies, and on generalisability (external validity). Based on the findings of this review, existing CPP clinical guidelines, textbooks on CPP and narrative reviews should be interpreted with caution. This because current recommendations for specific physiotherapy treatments are not evidence-based. That only small, single studies have been undertaken of most of these interventions greatly limits the available evidence on which clinical practice can be based. There seems to be some evidence to support Mensendieck somatocognitive therapy, and the use of a multidisciplinary intervention, but further work is required to confirm these findings.

4.5 Implication for further research

Given the prevalence and healthcare costs associated with female CPP methodologically robust primary research should be designed and conducted to test the effect of physiotherapeutic interventions for CPP. The clinically relevant levels of pain improvement demonstrated in the RCTs included in this systematic review should inform hypotheses for future high quality RCTs in order to reinforce the evidence base for physiotherapeutic interventions. The conduct of and critical appraisal of future RCTs should follow the CONSORT statement checklists and flow diagram [55], and refer to important outcome domains [56] and relevant length of treatment/time to follow-up. This would strengthen the provided evidence remarkably.