Dehydroepiandrosterone for systemic lupus erythematosus

Criteria for considering studies for this review

Search methods for identification of studies

Electronic searches 
 We searched the following databases: 
 * The Cochrane Library ‐ all sections including the Cochrane Central Register of Controlled Trials (Issue 2, 2006) 
 * MEDLINE (1966 to June Week 2 2006) 
 * PubMed (limited to records added in the 60 days previous to 27 June 2006) 
 * EMBASE (1980 to 2006 Week 25) 
 * Science Citation Index (1980 to June 2006) 
 * ISI Proceedings (1990 to June 2006)

Ongoing trials database 
 * Current controlled trials (searched 27 June 2006)

Search strategy 
 We ran the following search strategy in MEDLINE and adapted as appropriate for the other databases: 
 1. exp Dehydroepiandrosterone/ 
 2. (prasterone or dehydroepiandrosterone or DHEA or GL701 or Aslera or prestara or anastar).tw. 
 3. 1 or 2 
 4. exp Lupus Erythematosus, Systemic/ 
 5. (lupus or sle).tw. 
 6. 4 or 5 
 7. 3 and 6

There were no language restrictions on searching. 
 All results were downloaded and manually screened for RCTs.

Handsearches 
 We examined reference lists of the relevant trials and reviews identified for additional studies.

Additional searches 
 * The web site of Genelabs (http://www.genelabs.com) 
 * The web sites of the FDA and EMEA 
 * The web site of EULAR http://www.eular.org/ for meeting abstracts from 2002 to 2006. 
 * ADIS R&D Insight (27 June 2006).

Data collection and analysis

Trial selection 
 Two observers (CB, PR) independently reviewed titles, abstracts and key words of all the records retrieved in the search. Full articles were then retrieved for further assessment when the information given suggested that the study: 
 1. included patients with SLE; 
 2. compared DHEA to placebo; 
 3. assessed one or more of the defined outcome measures; 
 4. met the defined criteria for trial inclusion.

Full articles were retrieved for clarification when there was doubt about eligibility. Inter‐rater agreement was assessed using Cohen's kappa to be 1. Had differences of opinion existed, they would have been resolved by discussion with a third party (DC). If resolving any disagreement had not been possible, the article would have been added to those 'awaiting assessment' and the authors contacted for clarification. If no clarification was provided, the review group editorial base would be consulted. No papers underwent this process.

Quality assessment 
 The quality of reporting of each trial was based largely on the quality criteria specified by Schulz and Jadad (Schulz 1995; Jadad 1996) and as described in the manual of the Centre for Reviews and Dissemination (CRD) for randomised controlled trials (Khan 2000).

In particular the following factors were assessed: 
 1. Minimisation of selection bias 
 a) was the randomisation procedure adequate? 
 b) was the allocation concealment adequate?

2. Minimisation of attrition bias 
 a) were withdrawals and dropouts completely described? 
 b) was analysis by intention‐to‐treat?

3. Minimisation of detection bias 
 a) were outcome assessors blind to the intervention?

Based on these criteria, studies were broadly subdivided into the following three categories (see Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2006): 
 A ‐ all quality criteria met: low risk of bias. 
 B ‐ one or more of the quality criteria only partly met: moderate risk of bias. 
 C ‐ one or more criteria not met: high risk of bias.

This classification would have been used in the sensitivity analysis if appropriate. Two review authors independently assessed each trial. Inter‐rater agreement was calculated using Cohen's kappa to be 1. Had disagreement occurred, the rest of the group would have been consulted and a judgement made based on consensus.

Data extraction 
 Two review authors, using a predefined data extraction form, undertook data extraction independently. Differences were resolved by discussion, with reference back to the original paper, and if necessary, a third opinion was sought.

Extraction included the following information: 
 1. General information ‐ author and year, country, setting, published/unpublished, language of publication, source of funding, abstract / full article, duplicate publications. 
 2. Trial characteristics ‐ design, duration, quality by method and security of randomisation, blinding, allocation concealment. 
 3. Participants ‐ age, gender, race, selection method, exclusions; duration of SLE, initial composite scores, baseline similarity of groups (including co‐morbidity), assessment of compliance, other therapies. 
 4. Interventions ‐ type of DHEA, dose of DHEA, placebos 
 5. Results ‐ comparability at baseline, losses/drop‐outs, mortality and morbidity outcomes, adverse effects, patient preference, quality of life, study duration, as observed and by intention to treat.

Data analysis 
 Data were summarised statistically using meta‐analysis if they were available, sufficiently similar, and of sufficient quality. Dichotomous data were expressed as relative risks (RR). Continuous data were expressed as mean differences (MD) and an overall MD calculated. Overall results were calculated based on the fixed‐effect model. Heterogeneity was tested for using the Z score and the chi‐squared statistic with significance being set at P < 0.1. The I² statistic was used to estimate the proportion of the total variation in study estimates that could be explained by heterogeneity. If significant heterogeneity was found, it was considered to be unreasonable to assume that there was one 'true' effect underlying the data, that was constant across different populations, and therefore no summary statistic was calculated.

There were insufficient studies reporting data in a way that could be summarised to enable meaningful exploration of causes of heterogeneity. In the future, if more studies are published, we will assess possible sources of heterogeneity by subgroup and sensitivity analyses as described below. Had there been sufficient studies reporting data consistently, a funnel plot would have been undertaken to assess the effect of small studies on the estimates.

Subgroup analysis 
 We have undertaken subgroup analyses based on baseline disease severity. Only one small trial has been published in people with severe SLE. We therefore reported the findings for mild to moderate and severe separately.

We planned to perform the following additional subgroup analysis in order to explore effect size differences, had the data allowed. 
 1. Different types of DHEA analogues. 
 2. Dose. 
 3. Age. 
 4. Gender. 
 5. Race.

Sensitivity analyses 
 Had it been appropriate, the following sensitivity analyses would have been performed in order to explore the influence on effect size. 
 1. Repeating the analysis excluding studies published in abstract form only. 
 2. Repeating the analysis taking account of study quality, as specified above. 
 3. Repeating the analysis excluding any very long or large studies to establish how much they dominated the results. 
 4. Repeating the analysis excluding studies using the following filters: language of publication, source of funding (industry versus other), country if appropriate.

Clinical relevance tables 
 Clinical relevance tables were compiled under additional tables to improve the readability of the review. For any primary outcomes where a statistically significant difference between the treatment group and control group had been demonstrated and the outcome was dichotomous, like adverse events, the number needed to treat was calculated from the control group event rate and the relative risk using the Visual Rx NNT calculator (Cates 2003). For continuous primary outcomes, tables have also been presented under additional tables. When statistically significant treatment effects were demonstrated for any of the primary outcomes then the absolute benefit was calculated as the improvement in the intervention group minus the improvement in the control group, in the original units. Similarly relative difference in the change from baseline was calculated as the absolute benefit divided by the baseline mean of the control group. NNT was calculated using the Wells calculator software available at the CMSG editorial office. The minimal clinically important difference (MCID) for each outcome was determined for input into the calculator.

Grading of evidence 
 We used the grading system described in the 2004 book Evidence‐based Rheumatology (Tugwell 2004) and recommended by the Musculoskeletal Group:

Platinum: A published systematic review that has at least two individual controlled trials each satisfying the following: 
 * Sample sizes of at least 50 per group ‐ if these do not find a statistically significant difference, they are adequately powered for a 20% relative difference in the relevant outcome. 
 * Blinding of patients and assessors for outcomes. 
 * Handling of withdrawals > 80% follow up (imputations based on methods such as Last Observation Carried Forward (LOCF) are acceptable). 
 * Concealment of treatment allocation.

Gold: At least one randomised clinical trial meeting all of the following criteria for the major outcome(s) as reported: 
 * Sample sizes of at least 50 per group ‐ if these do not find a statistically significant difference, they are adequately powered for a 20% relative difference in the relevant outcome. 
 * Blinding of patients and assessors for outcomes. 
 * Handling of withdrawals > 80% follow up (imputations based on methods such as LOCF are acceptable). 
 * Concealment of treatment allocation.

Silver: A randomised trial that does not meet the above criteria. Silver ranking would also include evidence from at least one study of non‐randomised cohorts that did and did not receive the therapy, or evidence from at least one high quality case‐control study. A randomised trial with a 'head‐to‐head' comparison of agents would be considered silver level ranking unless a reference were provided to a comparison of one of the agents to placebo showing at least a 20% relative difference.

Bronze: The bronze ranking is given to evidence if at least one high quality case series without controls (including simple before/after studies in which patients act as their own control) or if the conclusion is derived from expert opinion based on clinical experience without reference to any of the foregoing (for example, argument from physiology, bench research or first principles).

Description of studies

The searches retrieved a total of 480 references. After the removal of duplicates there were 325 unique references. These contained seven RCTs, published in eight full papers (Petri 2004 comprised two separate publications), which met our inclusion criteria.

Excluded studies 
 Three full publications (van Vollenhoven 1994; Barry 1998; Chang 2004) and five abstracts (van Vollenhoven 1992; van Vollenhoven 1993; vanVollenhoven 1994b; van Vollenhoven 1996; van Vollenhoven 2001) were excluded from this review.

All apart from Chang 2004 and van Vollenhoven 2001 were case series and excluded on the basis of study design.

Chang 2004 was conducted as a sub‐study of Chang 2002 (a multi centre randomised placebo controlled trial over 24 weeks) and evaluated the role of cytokines in treatment of SLE with DHEA. It was excluded because this was not an outcome measures included in this review.

van Vollenhoven 2001 reported a double‐blinded placebo controlled trial assessing the effect of DHEA on cognitive function in participants with mild to moderate SLE over a one month treatment period. It was excluded from this review because it was less than three months in duration.

Included Studies 
 We identified seven studies for inclusion in this review. One of the studies had been reported in more than one full publication (Van Vollenhoven 1995) and several had been reported as a series of abstracts prior to full publication.

Study design 
 All seven included studies were RCTs of at least three months duration. All had been published in full. The shortest of the trials lasted three months (Van Vollenhoven 1995), three lasted for six months (Van Vollenhoven 1999; Chang 2002; Nordmark 2005), one for up to nine months (Petri 2002) and the longest trials studied people receiving DHEA for up to 12 months (Petri 2004; Hartkamp 2004). Two studies (Chang 2002; Petri 2004) included a six week period prior to participation in the study during which participants must have been stable on all existing therapy, and one study required an eight week stable period (Nordmark 2005). One study (Hartkamp 2004) required participants to be stable without steroids for at least six months. No specific justification was presented by the researchers for the 'run in' periods but in all cases, the study protocol specified that no change in other medication (other than the study drug) could occur during the trial. A further study (Petri 2002) required at least six weeks of stable prednisolone therapy, and four weeks stable on any other therapy. In this case the aim of the study was to reduce the prednisolone dose. 
 
 Participants 
 There were a total of 842 participants, of which 450 were on DHEA, but after the authors had excluded selected drop outs (two dropped out from Van Vollenhoven 1995; four dropped out from Nordmark 2005) 836 people were included in the analysis. Reflecting the epidemiology of SLE, most of the trials included only women. One study (Van Vollenhoven 1999) did not exclude male participants, but only 3 of the 19 participants were male. Caucasians were the most commonly represented racial group, although one trial included only Chinese women (Chang 2002). Where reported, the median age of trial participants was approximately 30 to 47 years. One study did not report the age of the study populations (Chang 2002).

Two studies restricted participation to people with active SLE as defined by a rheumatologist, using two disease activity scales (SLEDAI > 2 and SLAM >= 7 plus steroids use of less than 10 mg per day) (Chang 2002; Petri 2004). Van Vollenhoven 1995 defined participants as having "mild to moderate" disease but did not define this further. A further study restricted participants to those with SLE managed on less than 10 mg prednisolone (Hartkamp 2004). One study (Van Vollenhoven 1999) was restricted to people with severe SLE based on meeting the ACR criteria for severe renal, haematological or serosal disease. Petri 2002 only included people receiving 10 to 30mg of prednisolone per day. Nordmark 2005 limited the study population to those receiving at least 5mg of prednisolone daily.

Those with severe disease have been reported separately.

Interventions 
 DHEA was administered as a once daily oral dose in all but one (Nordmark 2005) of the trials. In one study, participants receiving active treatment were given 100 mg per day (Van Vollenhoven 1995). In four studies the active dose was 200 mg per day (Van Vollenhoven 1999; Chang 2002; Hartkamp 2004; Petri 2004). Petri 2002 included two active treatment arms with 100 and 200 mg daily doses in each (Petri 2002). Finally Nordmark 2005 administered low doses of DHEA (10 to 15 mg) twice daily to participants. All the trials included a placebo arm.

In total 450 people were exposed to DHEA. Seventy seven participants were treated with DHEA 100 mg daily. Three hundred and seventy three were treated with DHEA 200 mg daily. Three hundred and eighty six participants received a placebo.

Outcome measures 
 Primary outcomes 
 There is no single accepted measure of SLE activity and this was reflected in the range of measures reported in the seven included studies.

Disease Activity 
 SLEDAI was reported in six of the seven studies (Petri 2002 did not) and SLAM was reported in Van Vollenhoven 1999; Chang 2002; Petri 2004. Petri 2002 states that SLEDAI was measured but does not report data because "according to the trial design, a reduction in the dose of prednisolone at protocol specified visits was mandatory when a participant's SLEDAI score was stable or had improved. SLEDAI and other secondary outcome measures would not be expected to improve". 
 Three studies (Van Vollenhoven 1999; Petri 2002; Petri 2004) used "responder" measures as the primary outcome. In one (Petri 2004) a composite score was used based on improvement or stabilization of a number of predefined measures of disease activity. Petri 2002 defined responders to be anyone who achieved a reduction in prednisolone to 7.5 mg or less per day for two consecutive months. Van Vollenhoven 1999 defined responders as those meeting absolute predefined improvements in the main organ affected by their SLE.

The number of participants experiencing 'flares' of disease and the 'time to flare' were also reported as markers of disease activity (Van Vollenhoven 1995; Chang 2002; Petri 2004)

Damage 
 Only one of the studies used an index of irreversible end organ damage such as the American College of Rheumatology/ Systemic Lupus International Cooperating Clinics (SLICC) damage index (Nordmark 2005) reporting that it was unchanged in both the treatment and placebo groups but with no further details (one stated that it was measured but did not report any results Petri 2002). Some reported individual end organ damage (Van Vollenhoven 1999 reported renal damage) or included assessment of organ damage as part of their composite scoring index (Van Vollenhoven 1999; Petri 2004).

Health related quality of life 
 Only one of the studies reported health related quality of life using methods other than patient global visual analogue scales (Nordmark 2005). The authors reported a number of tools including Short Form‐36, a Hopkins Symptom Checklist (HSCL) and the psychological General Well‐Being index as well as a Swedish version of the McCoy Sex Scale Questionnaire. Four of the other studies reported patient global scores (Van Vollenhoven 1995; Van Vollenhoven 1999; Chang 2002; Petri 2004). The Patient global visual analogue scale is a simple generic tool allowing participants to indicate their perception of their health status on a 100 mm bar where 100 represents optimal and zero represents the worst possible. There are a number of disease specific questionnaires for use in SLE, including the SLEQoL and LupusQoL. None of these scales were reported. One study stated that the primary endpoint was well‐being and fatigue, but did not report any results for either of these endpoints (Hartkamp 2004).

Secondary Outcomes 
 The ability to reduce concurrent corticosteroid dose was assessed as the primary outcome in one study in the form of a "responder" definition (Petri 2002) but was also reported in two other studies (Van Vollenhoven 1995; Van Vollenhoven 1999). Changes in bone mineral density was measured in three studies (Van Vollenhoven 1999; Hartkamp 2004; Nordmark 2005) and for a subgroup of Petri 2004. Different summary measures were reported for each study.

Krupp fatigue severity score was reported in Petri 2004.

The 'Characteristics of included studies' table details the outcome measures included by each study.

Risk of bias in included studies

Details of the quality assessment is summarised in Table 1.

Only one of the trials was considered to fall into category A of the quality assessment detailed in the methods, minimising the risk of selection, attrition and detection bias (Petri 2004). Van Vollenhoven 1995, Petri 2002, and Hartkamp 2004 reported sparse (or no) details of the assignment and allocation concealment methods. Losses to follow up were poorly reported in two of the studies (Chang 2002; Petri 2002). Chang 2002 provided no information about the baseline characteristics of the study groups. Baseline characteristics in Van Vollenhoven 1999 showed some potentially important differences including the only male participants (N = 3) receiving placebo and substantial differences in the distribution of severely effected organs between treatment and placebo arms. Nordmark 2005 excluded four people who dropped out during the study period as a result of adverse events. It is not clear when these people withdrew from the study, but all four appear to have been in the active treatment group. We have retained these four people in the analysis of participants. Van Vollenhoven 1995 also excluded two people, one from each group. In this study withdrawal occurred before any treatment had been administered and therefore the authors decision to excluded them from analysis has been reflected in our analysis of participants.

In addition, two of the studies were very small, containing less than 30 participants (Van Vollenhoven 1995; Van Vollenhoven 1999).

Petri 2002 and Petri 2004 performed well against the quality criteria set out in the method. However, Petri 2002 undertook post hoc subgroup analysis, after unblinding, because initial analysis identified large number of participants experiencing a positive primary outcome (which included "stable disease") in both the treated and untreated groups. The researchers found that response correlated with SLEDAI score, with higher scores experiencing poorer responses in the placebo group. They presented results by subgroups of SLEDAI score at baseline. 
 Petri 2004 was complicated by a change in 'entry' requirements during the study period. The change, to include separate analysis for those with SLEDAI > 2, was added as a result of the findings of an earlier study (Petri 2002) which reported substantial differences in response among those with higher SLEDAI. The change to protocol was undertaken prior to the unblinding of the study. The authors, therefore, present two analyses for their primary outcome ‐ a composite measure; for all randomised participants, and for those meeting a stricter definition of 'active SLE' defined by SLEDAI > 2. For other outcome measures ‐ only those with SLEDAI > 2 were reported.

Effects of interventions

Primary outcomes 
 Disease activity 
 Of the six widely used disease activity measures only two were reported by any of the studies, the SLEDAI and the SLAM. The SLEDAI and SLAM are both global scoring systems which aim to provide an overall assessment of disease activity rather than being assessments of individual organ systems.

SLEDAI 
 The SLEDAI measures disease activity in the preceding ten days. It contains 24 weighted parameters and scores can range from zero to 105. Although it is possible to score 105 this is rarely seen in clinical practice, and a score of 20 represents very high disease activity. A SLEDAI score of 11 to 19 represents high disease activity, whilst 6 to 10 is moderate activity, and 1 to 5 is mild. Zero represents no disease activity.

Six of the studies reported disease activity using SLEDAI (Van Vollenhoven 1995; Van Vollenhoven 1999; Chang 2002; Hartkamp 2004; Petri 2004; Nordmark 2005). Four studies (Van Vollenhoven 1995; Van Vollenhoven 1999; Chang 2002; Hartkamp 2004) reported scores and one study (Petri 2004) reported, for the subgroup of participants who had active disease at baseline (defined as SLEDAI > 2), the percentage of participants who remained 'stable or improved'. One (Nordmark 2005) reported that there was no change in SLEDAI from baseline but gave no further details.

Mild to Moderate SLE 
 None of the studies reported statistically significant change in SLEDAI score from base line when comparing the treatment groups to the placebo group on an intention to treat basis. Three studies reported small changes in SLEDAI in the placebo and treatment arms (mean difference in change from baseline of less than 1 point on the scale) (Chang 2002; Hartkamp 2004; Nordmark 2005).

Meta‐analysis was possible for two of the studies (N = 148) (Van Vollenhoven 1995; Chang 2002) and supported no statistically significant mean difference in SLEDAI score among those treated with DHEA versus placebo (‐0.6; 95% CI ‐2.12 to 0.89). 
 Ranking of evidence: Gold based on the meta‐analysis of two studies (N = 148 with one RCT containing > 50 participants in each treatment group and the larger study (Chang 2002) being of good methodological quality). Three of the studies that could not be included in the meta‐analysis reported findings in keeping with the meta‐analysis result.

Petri 2004 reported 90.5% of participants experiencing a stabilisation or improvement in the SLEDAI score on treatment versus 82.2% of participants stabilising or improving without treatment (P = 0.04) if the analysis was restricted to those with active disease (SLEDAI > 2). This did not give any measure of the scale of change in SLEDAI detected in those who improved or those who deteriorated. Using the more lenient requirement of "stabilisation or improvement" in condition, Petri 2004 reported that a statistically significantly greater proportion of participants treated with DHEA met the criteria compared to placebo (8.3% stabilised or improved P = 0.04). This was gold ranking evidence.

Severe SLE 
 One small study, in people with severe disease, reported a mean improvement in SLEDAI score of 10.3 compared to an improvement of 3.9 in the placebo group (Van Vollenhoven 1999). (Number needed to treat (NNT)= 3) This was of borderline statistical significance, however, only 19 people participated in this study. 
 Ranking of evidence: silver

SLAM 
 The SLAM measures disease activity in the preceding month. The original index was published in 1989 and has been subsequently revised to the SLAM‐R. It measures 30 variables in 11 organ systems and 8 laboratory features. The minimum score is zero and the maximum is 84. A score of 7 is felt to indicate disease activity that would warrant modification of treatment in 50% of participants.

SLAM was reported in three studies (Petri 2004; Chang 2002; Van Vollenhoven 1999). Two of the studies reported scores (Van Vollenhoven 1999; Chang 2002) and one study (Petri 2004) percentage of participants who 'stabilised or improved' but only in a subgroup of those with active disease (SLEDAI > 2). None of the studies demonstrated a statistically significant change in the SLAM between the treatment and placebo groups. 
 Ranking of evidence: gold*

*Gold ranking evidence based on the meta‐analysis of two studies (N =148 with one RCT containing > 50 participants in each treatment group and the larger study (Chang 2002) being of good methodological quality). Petri 2004 reported findings consistent with no statistically significant difference in SLAM scores for those treated with DHEA as compared to placebo. 
 Table 2; Table 3; Fig 01,01 summarises the results.

Summary for Disease Activity : There was 'gold' level evidence and consistent findings across five studies that, in people with mild to moderate SLE, treatment with DHEA provided no benefit over placebo in terms of disease activity, measured by mean change in SLEDAI.

However, there was also 'gold' level evidence from a single study that, if only those with 'active' disease (SLEDAI > 2) were included in the analysis, 8.3% (P = 0.04) more people remain "stable or improve" on DHEA versus placebo, measured using SLEDAI .

This inconsistency in the evidence may arise because the criteria of "stabilised or improved" is considered to be a more lenient assessment than mean change in SLEDAI score. The inconsistency between findings does not seem to be explained by the restriction to only those with SLEDAI > 2 at baseline because the Chang 2002 study also included this restriction but found no evidence of improvement in SLEDAI score.

One very small trial in participants with severe SLE reported improvements in SLEDAI among those treated with DHEA that were of borderline statistical significance when compared with placebo.

There was consistent gold ranking evidence that DHEA did not provide clinical benefits when measuring disease activity using SLAM.

Organ damage 
 As SLE is a chronic multisystem disease it is important to assess chronic organ damage. The SLICC Damage index has been endorsed by the American College of Rheumatology and is the most widely used damage index. It assesses 41 variables in 12 systems and the damage must persist continuously for six months to score. The maximum possible score is 47, but it would be uncommon in clinical practice for participants to survive with scores = 12. Only approximately 50% of SLE participants would be expected to score on the SLICC/ACR damage index.

None of the trials presented data on SLICC/ACR damage index scores.

Health Related Quality of life 
 In any chronic disease the patient's perception of their health related quality of life is important. The most widely used survey is the generic SF36 (Short Form 36) which assesses eight domains of general health perception, physical and social functioning, limitation due to physical or emotional factors, mental health, vitality and pain. Lower scores represent poorer health status. Only one study reported SF36, while four reported Patient or Physician Global Assessment using a 0 to 100mm visual analogue scale. A clinically relevant change in Patient or Physician Global Assessment has been reported to be +/‐10mm.

SF‐36 
 Mild to Moderate SLE 
 Only Nordmark 2005 reported SF36, finding improvements in SF36 among those treated with low dose DHEA, when compared to those getting placebo, in the functional domains of role limitation due to physical factors and to emotional factors. This was of statistical significance in the domain of role limitation due to emotional factors (+ 23.3 versus ‐14.6). 
 Ranking of evidence: silver (small study of < 50 participants in each group)

Patient Global Assessment 
 Four of the studies reported the generic Patient Global Assessment (Van Vollenhoven 1995; Van Vollenhoven 1999; Chang 2002; Petri 2004). Another two studies measured but did not report Patient Global Assessment (Petri 2002; Hartkamp 2004).

Mild to Moderate SLE 
 Both Chang 2002 and Van Vollenhoven 1995 reported statistically significant improvements in the Patient Global Assessment in the treatment group compared to deterioration in the placebo (after adjustment for other study factors).

Meta analysis was possible for these two studies and supported a statistically significant mean difference in Patient Global Assessment (11.5% reduction (reduction of 11.5 mm; 95% CI ‐19.08 to ‐3.84) on a scale of 0 to 100 mm) ] in those treated with DHEA versus those receiving placebo. (NNT to achieve this level of improvement =5) 
 Ranking of evidence: Gold (N = 148 with one RCT containing > 50 participants in each treatment group and the larger study (Chang 2002) being of good methodological quality).

One study (Petri 2004) reported that, for the subgroup of participants with active disease (SLEDAI > 2) a statistically significant lower percentage of participants deteriorated in their Patient Global Assessment in the DHEA group (10.9%) compared to placebo (22.6%) (P = 0.007).

Severe SLE 
 One study (Van Vollenhoven 1999) found no difference in Patient Global Assessment between treatment and placebo groups. 
 Ranking of evidence: silver.

Table 4 and Fig 0201 summarises these findings.

Physicians Global Score 
 The Physicians global score was reported in three of the studies (Van Vollenhoven 1995; Van Vollenhoven 1999; Chang 2002). In none of the studies was a statistically significant improvement achieved.

Mild to moderate SLE 
 Meta‐analysis was possible for the two studies (Van Vollenhoven 1995; Chang 2002) and supported no statistically significant mean difference in Physician Global Score (‐3.2; 95% CI ‐8.1 to 1.8) for treatment with DHEA versus placebo. 
 Ranking of evidence: gold (N = 148 with one RCT containing > 50 participants in each treatment group and the larger study (Chang 2002) being of good methodological quality).

Hartkamp 2004 stated that the main outcome of interest in their study was wellbeing and fatigue. We can find no report of these endpoints in the published literature.

Summary of Health Related Quality of Life : There was 'gold' level and consistent evidence that treatment with DHEA did statistically significantly improve quality of life, measured by Patient Global Assessment, when compared with placebo. There was evidence of improvement in some, but not all, domains of the SF36 tool when treated with low dose DHEA. Using Physician Global Assessment, there was gold level and consistent evidence of no clinical benefit.

Adverse Events 
 Table 5 and Figs 0301 to 0303 summarises the adverse events 
 Adverse events were not consistently reported throughout the studies. It should be noted that two studies (Hartkamp 2004; Nordmark 2005) did not report any information about adverse events. Due to DHEA androgenic properties, we anticipated that certain adverse events might be reported (cardiovascular disease and hypertension, acne and hirsutism, cancer). These will be considered specifically below but general adverse event reporting is outlined first.

Withdrawal of treatment due to adverse events 
 Two studies reported withdrawals of treatment thought to relate to adverse events (Petri 2002; Petri 2004). Petri 2004 reported statistically significantly more withdrawals due to adverse events among those treated with DHEA (14.3% versus 5.7%; P = 0.005). Most of this difference was attributed to androgenic adverse events (acne and hirsutism). Similar rates of withdrawal in active and control groups were reported in the other study (Petri 2002) with 5% withdrawal in the placebo and 6 to 9% in the DHEA groups (two different drug doses). Van Vollenhoven 1995 noted that there were no withdrawals from treatment due to adverse drug reactions.

Serious adverse events 
 Four of the studies reported an overview of adverse events classed as 'serious' by the assessing physicians (Van Vollenhoven 1995; Van Vollenhoven 1999; Chang 2002; Petri 2004). Petri 2004 reported 14% of participants receiving DHEA as experiencing serious adverse events compared to 17% on placebo. Most of these (16/33 on placebo; 14/27 on DHEA) resulted in cessation of the study drug. No deaths were experienced in the DHEA treated group but five occurred among those receiving placebo (1 x pulmonary hypertension; 2 x suicide; 1 x sudden death; 1 x non‐Hodgkin's lymphoma). Van Vollenhoven 1999 reported only three serious adverse events: one death in the DHEA group from SLE complications shortly after randomisation; one deep venous thrombosis and one with fever associated with leucopaenia in the placebo group). In Chang 2002, 11.5% on DHEA experience serious adverse events, compared with 30.5% on placebo (P = 0.01) but this included events that were consistent with SLE disease flares. Van Vollenhoven 1995 noted that two serious adverse events had been reported, one in placebo and one receiving 200 mg DHEA. No further details were given. 
 NB no standard definition of "serious" was reported.

Cardiovascular events 
 None of the studies reported on the effect of DHEA on blood pressure. 
 Petri 2004 reported chest pain among 10.4% of those receiving DHEA versus 7.4% of those on placebo, a difference that was not statistically significant and no further details were given. None of the other trials reported on chest pain or any other cardiac events.

Acne, hirsutism and menstrual change (Figures 03/01; 03/02; 03/03) 
 Acne was the most commonly reported adverse event for DHEA occurring in between 33% and 59% of people receiving DHEA. That compares with 7 to 29% reporting acne in the placebo groups. Hirsutism, weight gain, and menstrual change where among the other androgenic side effects reported.

Mild to moderate SLE 
 Meta‐analysis of the four studies presenting data for acne in a suitable format (Van Vollenhoven 1995; Chang 2002; Petri 2002; Petri 2004;) estimated the relative risk of acne to be 2.16 (95% CI 1.65 to 2.83) in those treated with DHEA versus those on placebo. This difference was statistically significant. For hirsutism, the studies reported heterogeneous results, varying from no statistical difference to placebo to a relative risk of 10.2 associated with treatment. As no clear definition of hirsutism was given by any of the trialists, we have not presented a summary statistic for this adverse event. No statistical difference was reported for menstrual change (Relative risk 1.2; 95% CI 0.38 to 3.77). (Number Needed to Treat to obtain this level of acne adverse events = 5) 
 Ranking of evidence: Gold.

Severe SLE 
 From the one small study, no statistically significant difference in acne or hirsutism was reported between the treatment groups. Menstrual change was reported more often by those treated with DHEA (Relative Risk 2.67; 95% CI 0.98 to 7.22) (Van Vollenhoven 1999). 
 Ranking of evidence: Silver.

Cancer 
 Only one study reported identifying cancers (Petri 2004). All three tumours occurred in the population receiving placebo. One was a fatal non‐Hodgkin's lymphoma, the other two were carcinoma of the lung and breast.

Other reported adverse events 
 Petri 2004 provided the most comprehensive reporting of adverse events. Headache was reported in 56 of the 192 participants receiving DHEA (29.2% versus 22.2% placebo); 30 reported abdominal pain (25.6% versus 14.3%); 30 reported mood change (15.6% versus 14.8%); and 21 reported sinusitis (10.4% versus 9.0%). Musculoskeletal symptoms, rashes, fever, peripheral vascular disease and flu like symptoms were also reported among participants but are difficult to distinguish from disease activity.

Long term and rare adverse events 
 The longest exposure to DHEA in any of the randomised trials was 12 months (Hartkamp 2004; Petri 2004) but as noted above Hartkamp 2004 did not report on adverse events. Total exposed patient experience from the included RCTs was limited to 450 people who received active therapy.

Secondary outcomes 
 In addition to the primary outcomes above, we identified, a priori, the following secondary outcomes as of clinical relevance. These outcomes were reported inconsistently in all trials.

Specific Organ Damage Markers 
 One study (Van Vollenhoven 1999) presented data on change in proteinuria over six months, with reductions seen in both the placebo and DHEA group and no statistically significant difference between the groups.

Flares 
 Defining a flare in SLE is difficult as it can represent anything from major organ involvement to a change in the patient's social environment. Some investigators base the definition on change in the disease activity score whilst others require change in organ specific manifestations of the disease. Different definitions of a flare where used in each of the three studies reporting on flares making comparisons across studies impossible.

One study (Petri 2004) gave a five point criteria for defining a flare and, in addition, reported time to flare. One study (Chang 2002) modified the description of a flare from the SELENA study (FitzGerald 1999; Petri 2005) by increasing the glucocorticoid dose by 2.5 mg for at least seven days for SLE related reasons. These authors also reported time to first flare. One study (Van Vollenhoven 1995) assessed flares retrospectively based solely on whether the term "flare" was used in the patient's chart by the primary rheumatologist.

In all three studies (Van Vollenhoven 1995; Chang 2002; Petri 2004) a smaller percentage of participants flared in the DHEA group compared to the placebo group (Petri 2004 placebo 23.8% versus DHEA 29.7%, P = 0.27 (significant reduction if analysis restricted to those with active disease); 34% versus 18%, P not significant; 57% versus 21%, P = 0.053).

Fatigue 
 Fatigue can be a disabling symptom in SLE. Whilst fatigue has been shown to influence scoring on the SF36, there are specific tools available to assess fatigue. The Krupp Fatigue Severity Score was developed to assess fatigue in participants with Multiple Sclerosis and SLE. It is a nine items with a seven‐point response with high scores representing greater perception of fatigue. One study (Petri 2004) reported the mean change in score with 14.4% of participants on placebo deteriorating compared to 10.9% of those on DHEA; this was not statistically significant (only reported for those with active disease).

Steroid Use 
 Corticosteroid therapy has an important role in the management of SLE. Therapy may require to be initiated in response to an increase in disease activity or the dose of corticosteroid may require to be increased if the patient is already on maintenance corticosteroid therapy.

Three studies reported on corticosteroid use (Van Vollenhoven 1995; Van Vollenhoven 1999; Petri 2002). Three studies reported the mean change in corticosteroid use from baseline to final visit (Van Vollenhoven 1995; Van Vollenhoven 1999; Petri 2002) whilst one study (Petri 2002) also reported the number of days patient in the study were maintained on a dose of 7.5 mg or less of prednisone.

Change in steroid dose (Van Vollenhoven 1995; Van Vollenhoven 1999) showed a reduction in corticosteroid use in the DHEA group compared to placebo in all three studies, but none were statistically significantly different from the placebo group. In Van Vollenhoven 1999 a mean reduction of 36.2 mg was achieved in the DHEA group as compared with a reduction of 30.2 mg in the placebo group (P = 0.58). Van Vollenhoven 1995 reported a mean reduction of 3.2 mg in the DHEA group versus a 2.4 mg increase in the placebo group (P = 0.1). Petri 2002 reported a 36% reduction in mean prednisolone dose in the placebo group versus a 30% reduction in the 200 mg DHEA group (P = 0.67).

The one study (Petri 2002) which reported on the number of days participants were maintained on a dose of 7.5 mg or less of prednisone during the study period required participants to be on corticosteroid doses of between 10 to 30 mg of prednisolone for a period of 12 months prior to recruitment to meet the inclusion criteria. The study was designed with the aim of reducing steroid dose and if participants' disease was stable or improved, then steroids were reduced. The investigators found participants on placebo spent a median of 66.5 days on 7.5 mg or less of prednisolone. Participants on 100 mg of DHEA spent a median of 81 days on doses of prednisolone of 7.5 mg or less, whilst those on 200 mg of DHEA spent a median of 111.5 days on doses of 7.5 mg prednisolone or less. This finding was statistically significant for those on 200 mg DHEA versus placebo (P = 0.07). This study noted a high proportion of those with low SLEDAI scores managed to maintain prednisolone doses of less than 7.5 mg per day regardless of treatment.

Post hoc subgroup analysis of those with SLEDAI scores greater than 2 at baseline reported similar findings with a median of 28 days on a dose of 7.5 mg or less prednisolone for those on placebo (N = 45) and 110 days for those on 200 mg DHEA (N = 45) (P = 0.013).

Steroid Complications 
 Prolonged use of corticosteroids is clearly associated with corticosteroid related side effects such as osteoporosis, hypertension and diabetes mellitus. As participants are now surviving far longer with their disease the toxicity of therapies is becoming increasingly relevant. Awareness of this was one of the driving forces behind the development of the SLICC/ACR damage index. One of the points on the SLICC/ACR index is osteoporosis.

Bone mineral density (BMD) was reported in Hartkamp 2004. This was assessed at the lumbar spine in all participants at baseline and six months and twelve months. At baseline BMD was 0.987 gm/cm² in the DHEA group and 1.009 gm/cm2 in the placebo group. After 12 months there was no significant difference in BMD from baseline of in terms of change between the two treatment groups (DHEA at 12 months 0.992 versus placebo 1.009). Menopausal status and presence of other medications aimed at improving BMD appeared to impact on the effect of DHEA, but numbers in each group were small. No data was presented about the impacted base on steroid experience, but the authors state that there was no significant difference in BMD depending on whether currently on steroids or not. It should be noted that the primary outcome in this study was reported to be well‐being and fatigue. No data was presented for these outcomes.

Van Vollenhoven 1999 reported BMD in all 21 participants at six months and reported a significant reduction in BMD at the lumbar spine in those on placebo versus no significant change in the DHEA group (data presented in graphical form only).

From the 381 participants in Petri 2004, a subgroup of 55 participants continuously receiving steroids for at least six months had BMD measured at baseline and again at six months (a further 11 underwent baseline examination but did not attend for follow up because of withdrawal from the study ‐ they were not included in the analysis). The authors report that BMD was statistically significantly improved in the DHEA treatment group as compared to placebo at both the lumbar spine (DHEA: +1.7% versus placebo: ‐1.1%; P = 0.003) and hip (DHEA: 2% versus placebo ‐0.3%; P = 0.013)

Nordmark 2005 reported limited BMD results in their trial of low dose DHEA. They found no increase in BMD at either lumbar spine, hip or total body measured by dual energy x‐ray absorptiometry (DEXA) scanning in the group receiving DHEA but no other details were given.

Additional outcome measure 
 For completeness, we report the findings of three trials that used different composite clinical measures.

Assessment of disease activity in SLE is difficult both in clinical practice and in studies. A number of the studies (Van Vollenhoven 1999; Petri 2002; Petri 2004) have reported on the use of composite scores or responder indices based on a range of different measures in each trial.

Petri 2002 based their definition of responder on maintaining a dose of prednisolone of 7.5 mg per day or less for at least two months. All participants on the trial had to be on a dose of 10 to 30 mg of prednisolone at the outset of the trial. They did not report a statistically significant higher proportion of responders among those treated with DHEA, when compared to placebo (% responders: placebo ‐ 41%; 100 mg ‐ 44%; 200 mg ‐ 55% P = 0.1). Prior to unblinding the trial the baseline characteristics of the participants were reviewed and the responder rate recalculated on the basis of level of disease activity as assessed by the SLEDAI, with participants being stratified into mild disease (SLEDAI = 2) and more active disease (SLEDAI > 2). The more active group accounted for 72% of all participants. The reasoning behind this redesign of the protocol was the high rate of 'responders' in the group with low baseline activity on the SLEDAI regardless of whether on placebo or DHEA 100 mg or DHEA 200 mg. On re‐analysis of only those with a SLEDAI greater than 2, 29% of those on placebo "responded" versus 38% of those on 100 mg DHEA and 51% on 200 mg DHEA per day (200 mg versus placebo was statistically significant; P = 0.03).

One study (Petri 2004) a priori defined responders as participants who experienced no clinical deterioration and had improvement or stabilisation over the duration of the study in two disease activity measures (SLEDAI and SLAM) and two quality of life measures (patient's global assessment and KFSS). A total of 381 participants were recruited initially, but while the trial was progressing, data from an earlier trial resulted in a modification of the inclusion criteria (Petri 2002). While the trial was still blinded, a subgroup was identified as those with a SLEDAI > 2. Overall, no significant difference was reported in responder rates with 42% response among those receiving placebo versus 51% among those receiving DHEA. In the sub‐group with SLEDAI > 2 a statistically significant difference in percent of responders was reported; 45% responded in the placebo group and 59% in the DHEA group (P = 0.02).

Van Vollenhoven 1999 defined responders by stabilisation of the major lupus manifestation at the six‐month time point giving specific definitions for stabilisation for renal, haematologic and serositis manifestations. Seven of the nine participants receiving DHEA were classified as responders as compared to four of ten who received placebo.

Subgroup analysis 
 We identified six subgroups that would be of clinical importance in terms of exploring effect size differences for primary outcomes:

1. Type of DHEA analogue: Only one DHEA analogue was used in the studies reported to date so no subgroup analysis was appropriate

2. Dose: Only one study compared different doses of DHEA (Petri 2002). Sixty three participants received 100 mg DHEA and 64 received 200 mg DHEA. This study did not report any of the primary outcome measures we identified other than adverse events (Table 5). There were slightly more withdrawals due to adverse events as the dose increased (placebo: 5%; 100 mg: 6%; 200 mg: 9%). Androgen related adverse events (hirsutism, acne, menorrhagia) were not higher among those treated with 200 mg when compared to those receiving 100 mg doses. One other study used a 100 mg dose of DHEA (Van Vollenhoven 1995). This was a small study of 30 participants but it did demonstrate improvements in both SLEDAI and patient global scores. Androgenic adverse events were reported with acne being described in 57% of those receiving DHEA.

One study used the lower dose of DHEA of 10 to 30mg twice a day. It reported positive quality of life benefits from DHEA for some of the SF36 domains but no effect on disease activity.

3. Baseline disease severity: None of the trials allowed within study comparison of severity of disease. There is no universally accepted classification of severity and definition varied between trials. One study only included participants with severe disease and the results for this group have been presented separately Van Vollenhoven 1999.

4. Age: results were not reported by age bands therefore it was not possible to undertake any subgroup analysis.

5. Gender: Only one of the studies did not exclude males, however, only 3 of the 19 participants recruited were male. Results were not reported separately for the males, and all three received placebo.

6. Race: The predominant race reported by four of the trials (Nordmark 2005 did not report race) was Caucasian and none of these studies reported results separately for different ethnic groups (Van Vollenhoven 1995; Petri 2002; Hartkamp 2004; Petri 2004). The Van Vollenhoven 1999 was not predominantly Caucasian but little detail was given about race and the results were not reported separately. One study (Chang 2002) was conducted exclusively in Chinese women. This study did not report a positive benefit from DHEA in terms of disease activity (as measured by SLEDAI or SLAM) or quality of life. There may be some differences in the reporting or occurrence of adverse events that are related to race/culture. Very few androgenic adverse events were reported in this study, other than acne.

Sensitivity analysis 
 A priori we indicated four sensitivity analyses that would be relevant. There was very little overlap in the studies to enable meta‐analysis, either in terms of the disease severity or outcomes reported. For no outcome was it possible to include more than two trials in any given disease severity subgroup in a meta analysis. No sensitivity analysis was undertaken.

A summary of the meta‐analyses of benefits and harms are presented in clinical relevance tables (Table 6;Table 7).

Characteristics of included studies [ordered by study ID]

Characteristics of excluded studies [ordered by study ID]

Internal sources

• Department of Public Health, University of Aberdeen, UK.

External sources

• Health Technology Assessment program (via TAR budget), UK.