Three important points can be concluded from this systematic review of large scale, prospective, observational studies conducted in patients with psoriasis or psoriatic arthritis. First, very few large-scale, prospective, observational studies have been conducted given the burden of these diseases on society and the recent introduction of biologic agents onto the market, with only two assessing a drug versus drug comparison. Psoriasis is the most prevalent autoimmune disease in the United States. It affects 125 million people worldwide (2-3% of the total population). Between 10 and 20% of people with psoriasis will develop psoriatic arthritis. These conditions cause significant morbidity and have been associated with an increased risk of mortality compared to the general population. They significantly affect a patient's HRQOL and ability to carry out normal activities and the cost burden to society is substantial. In the United States psoriasis alone costs society $11.25 billion annually, with work loss accounting for 40% of this cost burden. The recent introduction of biological therapies represent an important addition to the approaches used in the treatment of psoriasis and psoriatic arthritis, however very few studies have assessed these agents in real-life situations compared to the more traditional treatments where many patient and provider factors, not present in clinical trial environments, can impact on effectiveness. Also, in some countries these agents are registered for use in specific target groups of patients where evidence of efficacy and safety are not provided by currently published clinical trials. Finally, clinical trial data only provide short-term evidence of efficacy and safety in a highly selected group of patients. For all these reasons large scale, long-term observational studies in real-life situations are needed to guide appropriate clinical and policy decision making.
Second, given the importance of collecting health economic data in a real world environment, very few observational studies collected data on economic outcomes or patient utilities. In the general hierarchy of clinical evidence in healthcare decision making, RCT's remain the gold standard for evaluation. However, there are a number of situations where such studies may be unnecessary, inappropriate, impossible or inadequate. The measurement of the effectiveness of a treatment, the longer-term outcomes of treatment (clinical and patient-reported), the true incidence of adverse events, and resource use associated with treatment and its side-effects are all situations where a RCT design is inadequate. RCT's often use patients, treatments and healthcare professionals that are all atypical and in addition are often short-term. Resource use and patient utilities observed in RCTs may not reflect that likely to be observed in regular clinical practice, not least because closer monitoring of patients in a trial may lead to events being detected and treated sooner than would otherwise be the case. This higher level of care may result in a small number of patients not experiencing high cost events that would be seen in everyday practice. In economic terms this is important since economic data is often highly skewed. The removal of a few observations with very high costs can have a large effect on overall health economic results. Also, RCTs are often conducted in specialist centres. The recorded resource consumption seen in the trial will therefore reflect the practice policies of this particular health care setting which may be very different to usual clinical practice. It is in such situations that observational cohort studies would provide more appropriate and informative health economic information if conducted and analysed rigorously.
Third, of those studies included in this review overall quality assessment was in general satisfactory, however the majority of studies failed to take into account and adjust for potential biases caused by lack of randomisation. Studies scored poorly on describing potential selection biases, identifying a comparison group, adjusting for confounders and losses to follow-up and providing adequate sample size calculations. The key question posed in cohort studies is the comparison of outcomes between two groups of patients (e.g. those responding to treatment vs. those not responding to treatment). Just over 60% of the studies in this review actually defined a comparison group, be it the general population or a more restricted internal or external population. For those studies not providing a comparison it is almost impossible to assess whether the results occurred by chance. Of those reporting a comparison group most studies reported potential selection bias however only half accounted for confounders and only a third accounted for losses to follow-up. In those studies not addressing these issues of potential bias, results are likely to have very low internal validity. Adjusting for the potential bias caused by lack of randomisation is critical to the validity of cohort studies[59–61].
When interpreting the results of this systematic review it is important to note three issues. First, it is difficult to systematically search for observational studies as search strategies that are both sensitive and specific do not exist for the major electronic databases. To overcome this problem we conducted a wide search and hand-searched reference lists of key papers. Second, consensus guidelines on the reporting of observational studies (STROBE) have only recently been introduced, therefore for many studies published prior to these guidelines it is often difficult to identify if the paper is a true observational study or not. Many studies stated they were observational, but in actual fact incorporated an experimental or 'open-label' element to them. Third, the cut-off of 100 patients to define large scale may have meant other important observational studies were excluded. However, only one study was excluded on the basis of sample size .
Large scale, prospective cohort studies are not the only non-randomised method for capturing real world health economic data. They are however, if conducted rigorously one of the best approaches to use, especially for non-rare outcomes over a relatively short period. A number of cross-sectional and case-control studies assessing cost, effectiveness and HRQOL have been conducted in patients with psoriasis and psoriatic arthritis. Cross-sectional studies are useful for assessing prevalence and describing specific characteristics of the disease, for example clinical and demographic characteristics, patient and provider perceptions of effectiveness, tolerability and compliance. However, unless they incorporate a retrospective element into their design, they are unable to distinguish between cause and effect and therefore are inappropriate for the measurement of effectiveness and health economic outcomes associated with an intervention. Retrospective elements to observational studies, for example retrospectively identifying patients or retrospective data collection (as in case control studies) introduces an additional level of bias and are therefore often used for more descriptive studies or hypothesis generation that can then be studied in a prospective observational study.
Looking outside of true observational designs to studies which are non-randomised but incorporate an experimental element to them, we find a number of 'open-label' trials, some aiming to assess longer-term outcomes and others aiming to assess effectiveness in a more naturalistic setting. These studies are not observational, although many claim to be. They are experimental in that patients have been selected for inclusion into the trial and administered the trial treatment. Given that these studies are not governed by any consensus guidelines on reporting or quality control, the potential for risk of error or bias is high and the results should be interpreted with caution. Included in these designs are 'open-label' extension studies. Patients represent a highly select group that have not only been selected on the basis of the original RCT, but are also those who have completed the randomized element to the trial and agreed to participate in the extension study. Such selection processes not only introduces significant bias, but also lowers even further the generalisability of the results to a wider population. In such studies the use of inferential statistics to allow for the possibility of sampling or random error to be the reason for the observed difference is crucial. However, in most extension studies assessing effectiveness in psoriasis or psoriatic arthritis no such inferential statistics have been carried out[62–64]. Also included are 'open-label' studies which adopt a non-randomised approach from the start of the study. Again these studies should be interpreted with caution for two main reasons; first, treatment is experimental and has therefore been selected by an investigator not independent from the study and second, the patient will know which treatment they are being given. Both actions will introduce inadvertent bias into the outcome assessment. Furthermore it is essential that such studies conform to the same rigorous methods expected of true observational studies in that the bias created from non-randomisation should be defined, explored and adjusted for. Currently, apart from one 'open-label' phase IV study assessing health economic outcomes which does account for confounding, most of the others don't[29, 36, 41].