Types of studies

RCTs (published or unpublished). Trials where pseudo‐randomisation methods are used, such as alternation, will be included.

Types of participants

Children and adults with defined CF, diagnosed clinically and by sweat test or genetic testing, including all ages, all degrees of severity of disease and any degree of liver involvement.

Types of interventions

UDCA administered orally, at any dose, given for a period of at least three months compared to a control group receiving either placebo or no additional therapy (i.e. both groups receiving usual CF therapy).

Types of outcome measures

Electronic searches

Relevant trials were identified from the Group's Cystic Fibrosis Trials Register using the terms liver AND ursodeoxycholic acid.

The Cystic Fibrosis Trials Register is compiled from electronic searches of the Cochrane Central Register of Controlled Trials (CENTRAL) (updated each new issue of the Cochrane Library), weekly searches of MEDLINE, a search of Embase to 1995 and the prospective handsearching of two journals ‐Pediatric Pulmonology and theJournal of Cystic Fibrosis. Unpublished work is identified by searching the abstract books of three major cystic fibrosis conferences: the International Cystic Fibrosis Conference; the European Cystic Fibrosis Conference and the North American Cystic Fibrosis Conference. For full details of all searching activities for the register, please see the relevant section of the Cochrane Cystic Fibrosis and Genetic Disorders Group's website.

Date of the most recent search of the Group's Cystic Fibrosis Trials Register: 09 April 2017.

The authors searched clinicaltrials.gov (https://clinicaltrials.gov/) and WHO ICTRPhttp://apps.who.int/trialsearch/ using the terms 'cystic fibrosis' AND 'ursodeoxycholic acid'.

Date of the most recent search: 13 April 2017.

Searching other resources

Reference sections of any trials identified were checked for any further RCTs. In addition we undertook full text searching of theJournal of Pediatrics from 1988 to 1995. We also contacted the pharmaceutical companies that market UDCA: Hoechst Marion Roussel (Destolit^®) and Consolidated Chemicals Ltd (Ursofalk^®).

Selection of studies

The two authors (KC and RS) independently applied the inclusion criteria to all potential reports.

Data extraction and management

We attempted to extract data from each RCT from the text, tables and figures. We recorded data on the number of participants with each outcome event, by allocated group, irrespective of compliance and whether or not the participant was later thought to be eligible or otherwise excluded from treatment or follow up. For continuous outcomes we recorded the mean change from baseline for each group and standard error or standard deviation.

Assessment of risk of bias in included studies

In order to assess the risk of bias in the included trials, we considered such aspects as generation of randomisation sequence and allocation concealment. If we regarded these as adequate then there was a low risk of bias to the trial; if we regarded these as inadequate, then there was a high risk of bias to the trial; and if they were considered unclear then the risk of bias was unclear too. We also considered the degree of blinding and the risk of bias increased as the number of people blinded to the intervention decreased. We also considered other risks of bias, e.g. from selective reporting.

Measures of treatment effect

We calculated a pooled estimate of the treatment effect for each outcome across trials. For binary outcomes we calculated, where possible, the odds of an outcome among treatment‐allocated participants to the corresponding odds among controls. For continuous outcomes, where data were available, we calculated a pooled estimate of treatment effect by calculating the mean difference (MD) and 95% confidence intervals (CIs).

Unit of analysis issues

Although we did not specifically excluded cross‐over trials, we were concerned about the use of a cross‐over design. This was because there may be a carry‐over effect of UDCA in the control arm. We did include one cross‐over trial in this review (Merli 1994). The data presented in the published report appeared to be combined from both treatment periods, but the authors attempted to overcome a possible carry‐over effect of UDCA by using a one‐month washout period. However, we considered it appropriate to compare only the first six months of the trial, i.e. UDCA versus placebo. Data from the first period were not available in the published report but the authors kindly provided the raw data. Including data from the first period in cross‐over trials in meta‐analyses is not without problems. Excluding later periods loses some of the information collected. Furthermore, if data from the first period are available in published reports they are likely to represent a biased subset of trials, usually because the authors have found evidence of carry‐over (Elbourne 2002).

Dealing with missing data

Where sufficient data were not available in the published reports or the abstract of the conference proceedings, the review authors attempted to contact the first and last authors of the paper.

We recorded data on the number of participants with each outcome event, by allocated group, irrespective of compliance and whether or not the participant was later thought to be eligible or otherwise excluded from treatment or follow up. This approach permits an intention‐to‐treat analysis.

Assessment of heterogeneity

We tested for heterogeneity between trial results using a standard Chi² test.

Data synthesis

We analysed the data using a fixed‐effect model. If, in future updates of this review, we identify a moderate to large degree of heterogeneity, we will analyse the data using a random‐effects model.

Subgroup analysis and investigation of heterogeneity

For future updates, if heterogeneity is identified and there are sufficient trials included in the review, we plan to investigate heterogeneity by means of examining individuals with evidence of liver disease at randomisation separately from those without liver disease.

Sensitivity analysis

We will also examine the robustness of our results using a sensitivity analysis including and excluding trials with a high risk of bias.

Summary of findings and assessment of the certainty of the evidence

In a post hoc change in line with current Cochrane guidance, at the 2017 update we added a summary of findings table (Table 1). We selected the following seven outcomes to report (chosen based on relevance to clinicians and consumers).

1. Change of hepatocellular enzymes from outside the normal range on at least one occasion to within the normal range of the method stated

2. Abnormally large livers reduced to within normal limits, as measured by ultrasound

3. Need for liver transplantation

4. Mortality

5. Nutritional status

6. Development of portal hypertension

7. Improved abnormal biliary excretion

We determined the quality of the evidence using the GRADE approach; and downgraded evidence in the presence of a high risk of bias in at least one trial, indirectness of the evidence, unexplained heterogeneity or inconsistency, imprecision of results, high probability of publication bias. We downgraded evidence by one level if they considered the limitation to be serious and by two levels if very serious.

Results of the search

Twelve trials have been identified as potentially relevant. Four trials were included (Colombo 1996;Lepage 1997;Merli 1994;O'Brien 1992) and eight trials were excluded (Bittner 1989;Colombo 1992;Kapustina 2000;Narckewicz 1994;NCT00004315;NCT00004441;Spray 2000;Van de Meeberg 1997).

Included studies

Four trials meet the inclusion criteria (Colombo 1996;Lepage 1997;Merli 1994;O'Brien 1992). One included trial is of cross‐over design but from the full published paper it is unclear whether there was any washout period employed and furthermore, data are not published for the first six‐month period of the trial; thus we are unable to extract appropriate data for analysis (Lepage 1997). As such, we are only able to present results from three trials (Colombo 1996;Merli 1994;O'Brien 1992).

Four trials with a total of 137 participants are included in the review, but results are only available from three trials involving a total of 118 participants (Colombo 1996;Merli 1994;O'Brien 1992). The ages of the participants ranged from 4 years to 32 years. The dose of UDCA given ranged from 10 to 20 mg/kg/day.

In three trials the comparison was with placebo (Colombo 1996;Lepage 1997;Merli 1994). In the fourth trial the comparison was with existing conventional therapy (O'Brien 1992). In three of the trials all of the participants had liver disease (Colombo 1996;Lepage 1997;O'Brien 1992), whereas in the third trial only 10 out of 51 participants had liver disease (Merli 1994).

The length of follow‐up was generally short and ranged from six months (Merli 1994;O'Brien 1992) to 12 months (Colombo 1996;Lepage 1997).

Important long‐term outcomes such as death or the need for liver transplant were not reported. Only two of our protocol‐defined outcomes were assessed: the nutritional indices (weight gain and skinfold thickness (Merli 1994;O'Brien 1992)); and biliary excretion (O'Brien 1992).

Trial design was complicated in three trials (Colombo 1996;Lepage 1997;Merli 1994).

In the Merli cross‐over trial, 51 participants were randomised to receive UDCA alone or with taurine for six months and then each treatment group was compared with a six‐month placebo period (Merli 1994). The sequence of treatment and placebo was then randomised in a cross‐over design. The data presented in the published report of the cross‐over RCT appeared to be combined from both treatment periods. Although we had not specifically excluded cross‐over trials, we were concerned about the use of a cross‐over design. This was because there may be a carry‐over effect of UDCA in the control arm, although the authors attempted to overcome this by using a one‐month washout period. However, we considered it appropriate to compare only the first six months of the trial, i.e. UDCA versus placebo. Data from the first period were not available in the published report but the authors have kindly provided the raw data. Including data from the first period in cross‐over trials in meta‐analyses is not without problems. Excluding later periods loses some of the information collected. Furthermore, if data from the first period are available in published reports they are likely to represent a biased subset of trials, usually because the authors have found evidence of carry‐over (Elbourne 2002).

In a further cross‐over trial we were unable to ascertain whether a washout period was employed and data were not presented for the first treatment period; we have not been able to clarify this information and have therefore decided not to present any results from this trial (Lepage 1997).

A factorial parallel design was employed in the Colombo trial (Colombo 1996). In this multicentre trial, 55 participants were randomised to receive UDCA or placebo and then each group was further randomised to receive either taurine or a second placebo. In effect, four parallel groups were studied.

In the O'Brien trial, 12 participants were randomised to UDCA or no additional therapy for six months (other than usual CF treatments, such as pancreatic enzymes and oral calorie supplements, which the UDCA group also received) (O'Brien 1992). Advanced liver disease, as documented by portal hypertension or histological features of fibrosis or cirrhosis or all three, was present in 11 out of 12 participants.

The use of taurine in two trials also complicated their design and analysis since taurine may affect liver involvement in CF (Colombo 1996;Merli 1994). Although UDCA is known to cause taurine depletion, the combined effect of UDCA and taurine on liver function is unknown.

These possible interactions and the complex trial designs caused difficulties when we considered combining the data. We have used subsets of the sample sizes given in the 'Characteristics of included studies' table so the participant numbers evaluated in the data tables and figures do not always tally with the sample sizes. In the Merli cross‐over trial we decided only to use data from the first six months of the UDCA/placebo group and not use the UDCA plus taurine group (Merli 1994). This gave us data on an unbalanced number of participants in the two groups in the first six‐month period: only six participants in the UDCA group and 12 participants in the control group (51 were initially randomised). In the factorial, parallel trial we decided not to use data from participants who received taurine (hence the total number of participants used in the data tables was 28 not 55) (Colombo 1996). However, only four of these 18 participants in this subset had abnormal liver enzymes at baseline.

Another issue of the Merli cross‐over RCT was that although weight, height and body mass percentile were measured, we had concerns about this type of trial design (Merli 1994). We would expect there to be a period effect on variables such as weight and height and this would require more subtle analysis. Again, we decided to use data only from participants in the UDCA‐alone group and from the first six months of the trial before cross over.

In 2005, Colombo presented follow‐up survival data (obtained by a data collection form sent to each centre) from the RCT that had been conducted in 1990 (Colombo 1996). Information was obtained from 53 of the original 55 participants (two were lost to follow‐up) for a median total period of follow‐up of 13.6 years; follow‐up data for the whole cohort were presented, not by randomised group. The majority of the trial participants had continued open UDCA therapy after the end of the trial (median daily dose 666 mg).

Excluded studies

Eight trials were excluded in total. Three trials were excluded because they did not include a placebo arm or a 'no UDCA' arm (Colombo 1992;NCT00004441;Van de Meeberg 1997). One trial was not an RCT (Narckewicz 1994). Two trials were of insufficient duration; in one the duration of follow up was only six weeks (Bittner 1989) and in the final cross‐over trial each treatment arm lasted four weeks (NCT00004315). Two trials were only published as abstracts (no full papers) with insufficient detail to confirm they meet the inclusion criteria; given the age of the abstracts, it is unlikely that any further publications relating to this trial will be forthcoming and therefore the trials have been excluded (Kapustina 2000;Spray 2000).

Allocation

All four trials were described as randomised, but only one trial stated the method used (Colombo 1996). We therefore judged the Colombo trial to have a low risk of bias (Colombo 1996), and the remaining three trials to have an unclear risk of bias (Lepage 1997;Merli 1994;O'Brien 1992).

Two trials described how allocation was concealed and these were judged to be adequate and hence have a low risk of bias (Colombo 1996;O'Brien 1992). The remaining two trials did not discuss allocation concealment and so were judged to have an unclear risk of bias (Lepage 1997;Merli 1994).

Blinding

One of the trials was described as double‐blinded and we judged this to have a low risk of bias (Colombo 1996). In the Merli trial, glucose tablets were used as the placebo, so it is probable, although not explicitly stated, that the participants at least were blinded to whether they were in the treatment or control group, due to this uncertainty we judged this to have an unclear risk of bias (Merli 1994). The Lepage trial did not describe the placebo or any other aspect of blinding so it was not possible to ascertain whether the participants or the trial personnel were blinded to treatment groups and this trial was also judged to have an unclear risk of bias (Lepage 1997). It was not possible to blind the O'Brien trial to participants or clinicians since the participants either received UDCA or no additional treatment, it was not discussed whether outcome assessors were blinded; and we therefore judged this trial to have a unclear risk of bias (O'Brien 1992).

Incomplete outcome data

An intention‐to‐treat analysis was performed in two trials (low risk of bias) (Colombo 1996;O'Brien 1992). In one trial 51 participants were initially recruited, but nine subsequently withdrew (Merli 1994). These participants were not followed up and were not included in the analysis; data from a further two participants were identified as being lost when the raw data were provided (high risk of bias). In the fourth trial six out of 19 participants withdrew; reasons were given for all six (unclear risk of bias) (Lepage 1997).

Other potential sources of bias

In the Colombo trial the characteristics of the two groups were not equal at baseline; the paper states that all five participants with oesophageal varices and seven out of eight participants with abnormal serum bilirubin levels at entry were allocated to the UDCA group (high risk of bias) (Colombo 1996). One cross‐over trial does not clearly report whether there was any washout period and data are not published for the first six‐month period of the trial (unclear risk of bias) (Lepage 1997). For the remaining two trials no other potential sources of bias were identified (low risk) (Merli 1994;O'Brien 1992).

Primary outcomes

Secondary outcomes

Although UDCA is relatively inexpensive compared to other CF treatments, it would need to be taken on a long‐term basis if it is effective. If it is ineffective then the resources saved by not using it could be used for other aspects of CF care.

Evidence of the effectiveness of UDCA is inconclusive. Routine use of UDCA in people with CF cannot, therefore, be justified. However, in view of these important preliminary results and because of the lack of any other effective intervention to prevent or treat CF‐related liver disease, it is essential that a large multicentre RCT of UDCA in people with CF is undertaken.

The results of this systematic review indicate that there is an urgent need for a well‐designed, adequately powered, multicentre RCT assessing the effectiveness of UDCA by measuring clinically relevant end points over years rather than months. Ideally a parallel trial, not a cross‐over trial, should be undertaken. However, as there is insufficient evidence to indicate that UDCA is effective in CF, it is not possible to suggest how long it should be given, but long‐term end points need to be assessed. Future trials should define the target population clearly, with separate trials for those without clinically detectable liver disease (the preventative effect) and those with liver disease (the therapeutic effect). In view of the problems of defining and assessing progression of liver involvement as well as the problem that by the time liver involvement is detected, it is too advanced for treatment, we suggest that the former is carried out first.

Although we have not been able to perform a formal meta‐analysis, the RCTs we have identified show important preliminary results.

Colombo 1996.

Lepage 1997.

Merli 1994.

O'Brien 1992.

Colombo 1996 {published and unpublished data}

Lepage 1997 {published data only}

Merli 1994 {published and unpublished data}

O'Brien 1992 {published and unpublished data}

Bittner 1989 {unpublished data only}

Colombo 1992 {published and unpublished data}

Kapustina 2000 {published data only}

Narckewicz 1994 {unpublished data only}

NCT00004315 {published data only}

NCT00004441 {published data only}

Spray 2000 {published data only}

Van de Meeberg 1997 {published data only}

Beuers 1992

Bhardwaj 2009

Carty 1995

CF Trust 2011

CF Trust 2016

Debray 2011

Elbourne 2002

Erlinger 1990

Frisancho 1990

Leeuwen 2014

O'Connor 1996

Parisi 2013

Poupon 1991

RLCH 2003

Roda 1982

Sokol 1999

Tanner 1992

Cheng 1997

Cheng 1999

Cheng 2012

Cheng 2014