Natural history studies provide invaluable data on the disease course. Rungoe et al1 report changes in medical treatment and surgery rates over the past three decades (1979–2011) using a large Danish population-based cohort of 48 467 patients with IBD issued from a Danish health administrative database. Interestingly, the authors found a decrease in surgery rates in both Crohn's disease (CD) and UC. Changes in surgery rates were accompanied by a significant decrease in the use of five aminosalicylic acids (5-ASA) and corticosteroids. The use of 5-ASA significantly decreased from 51% in 1995–2002 to 19% in 2003–2011 in CD; surprisingly, it also decreased in UC from 63% to 51%, while in UC 5-ASA remain the backbone therapy for mild-to-moderate UC according to recent international guidelines.2 As expected, there was a dramatic increase between 1995–2002 and 2003–2011 in the use of thiopurines and anti-tumor necrosis factor (TNF) 10 years after diagnosis: from 32% to 39% for azathioprine and from 10% to 23% for anti-TNF in CD. In UC, this increase was smaller than in CD and did not reach statistical significance. These rates are much lower than those reported in a French referral centre where the cumulative probabilities of receiving thiopurines and anti-TNF 5 years after CD diagnosis were 70% and 65%, respectively.3 Such discrepancy between referral centres and population-based studies demonstrates the importance to conduct cohort studies at the population level to avoid selection bias from referral centres in which patients with more severe disease are usually treated.

Overall, the results of this Danish study1 are very important in terms of public health and health economics. These results confirm recent data coming from a review article and a meta-analysis of population-based studies showing a decreased risk of intestinal surgery among patients with IBD over the past decades.4 ,5 However, as discussed by Runtgoe et al1 a causal link between this increased use of immunosuppressive therapy and the decreased need for surgery cannot be established due to potentially confusing factors. First, the impact of treatment duration on disease outcomes could not be evaluated in this Danish study1 and it is mandatory to initiate population-based, matched-cohort studies that used both time-dependent models and propensity score analysis.6 ,7 Hence, the authors arbitrarily predefined three groups according to the treatment duration of use (3–11 months, 12+ months and 3+ months) converting a continuous variable into a binary variable. Thereby, they did not perform a time-dependent analysis. Furthermore, they used propensity score expressed as likelihood of being treated with a given IBD-specific medication to reduce potential bias. Despite the broad utility of propensity score methods, when addressing causal questions from non-randomised studies, propensity score methods can only adjust for observed confounding covariates and not for unobserved ones.8

As for cohort studies in general, data collection in population-based studies must rely both on linkage with available databases (eg, hospital records, death certificates and registries) and on procedures designed especially for ascertaining study variables. The need for high quality, standardised data collection procedures makes quality assurance and quality key activities in population-based cohort studies, particularly those including multiple sites,9 and thus multiple data collectors. Health administrative databases are a large repository of all healthcare encounters for every legal resident and have been used to develop surveillance programmes for multiple chronic diseases, including diabetes10 and asthma.11 These population-based cohorts have been used to assess epidemiology, health services use and outcomes.10–12 The major weakness of the use of these administrative data to perform epidemiological studies is the difficulty to certify the exhaustively ascertainment, the definite IBD diagnosis and to obtain phenotypic data. The development and rigorous validation of an algorithm by Benchimol et al13 in paediatric-onset IBD cohort from Ontario has allowed to accurate identification of a population-based cohort using health administrative data from multiple sources within a universal access health system.

The Danish study overcomes most of these limitations by its powerful, nationwide sample size and impressive person-year follow-up.1 However, the absence of adjustment on disease pattern may have introduced a bias. In order to minimise bias, prospective well-designed cohort studies designed to specifically assess the impact of IBD-related medications on the natural history of the disease are eagerly awaited. Recent population-based studies prospectively collecting phenotype data could demonstrate that reduction in surgical rates was associated with increased and earlier azathioprine use, even though sample size might be an issue with such studies.6 ,7 Unfortunately, from a nationwide health administrative database it is not possible to obtain reliable disease phenotype data. Randomised controlled trials are usually underpowered and the duration of follow-up is too short to assess such issue. The Inflammatory Bowel South-Eastern Norway (IBSEN) cohort, a prospective population-based cohort study from Norway provided unique information on the natural history of IBD but was conducted in the prebiologic era.14 Paediatric-onset and elderly-onset population-based cohorts issued from the Epimad's Registry in Northern France also provided robust data on the course of IBD at the population level in France,15 while the Cancers Et Surrisque Associé aux Maladies inflammatoires intestinales En France (CESAME) cohort study could demonstrate an increased risk of lymphoma and non-melanoma skin cancer in thiopurine-treated patients before the advent of anti-TNF therapy.16 ,17 In the study by Rungoe et al,1 the use of oral corticosteroids was associated with a reduced risk of surgery. This finding is quite surprising as corticosteroids are not maintenance agents and do not have the potential for disease modification in IBD; this finding highlights the limitations of retrospective cohort studies using registries without adjustment on disease activity and severity. Upcoming prospective cohort studies on the natural history of IBD in the biologics era should include a regular assessment of objective signs of inflammation such as C reactive protein, faecal markers, endoscopy and/or cross-sectional imaging in addition to data on hospitalisations and surgeries.

Importantly, in the Danish study, the primary endpoint was surgery.1 Is preventing intestinal resection the ultimate goal of IBD treatment? The need for surgery in IBD is often considered as an objective disease outcome, mirror of the natural history and easily evaluated in retrospective studies. However, surgery can be a good option in some cases such as short ileal stenosis accompanied by obstructive symptoms. Hence, it is anticipated that the need for surgery will not disappear with a wider and earlier use of biologics as intestinal resection will remain in some cases an appropriate therapeutic choice for IBD. New therapeutic goals such as prevention of disability and bowel damage have emerged in IBD. These new endpoints should be used in future disease-modification trials to assess the real impact of IBD-related medications on disease course and replace surgery as a primary endpoint.18

In conclusion, all these data underscore the difficulty to conduct a good epidemiological approach to study the disease course and the impact of current therapeutic strategies in a chronic condition such as IBD. A first real disease-modification trial, Randomised Evaluation of an Algorithm for CD (REACT-2), is currently investigating whether an accelerated step-up treatment aiming to achieve and maintain deep remission leads to less CD-related complications as compared with the classical step-care approach. This trial will provide us unique information on the potential of thiopurines and biologics for disease modification in IBD.