- Email: [email protected]
Seasonal variation in flares of inflammatory bowel disease
GASTROENTEROLOGY 2004;126:665– 673
Seasonal Variation in Flares of Inflammatory Bowel Disease JAMES D. LEWIS,*,‡,§,㛳 FATEN N. ABERRA,‡ GARY R. LICHTENSTEIN,‡ WARREN B. BILKER,*,§,㛳 COLLEEN BRENSINGER,*,§ and BRIAN L. STROM*,‡,§,㛳,¶ *Center for Clinical Epidemiology and Biostatistics; ‡Department of Medicine; §Department of Biostatistics and Epidemiology; 㛳University of Pennsylvania Centers for Education and Research on Therapeutics; and ¶Department of Pharmacology, University of Pennsylvania School of Medicine, Philadelphia, Pennsylvania
Background & Aims: Previous research has yielded conflicting data as to whether the natural history of inflammatory bowel disease follows a seasonal pattern. The purpose of this study was to determine whether relapse of inflammatory bowel disease follows a seasonal pattern either across a cohort of patients or within individual patients. Methods: We used 1988 to 1997 data from the General Practice Research Database to conduct a retrospective cohort study of 1587 patients with Crohn’s disease (mean age at start of follow-up, 41 ⴞ 17 years) and 2773 patients with ulcerative colitis (mean age at start of follow-up, 48 ⴞ 16 years). Flares of disease were identified by receipt of a new prescription for either corticosteroids or 5-ASA medications following an interval of at least 4 months without prescriptions for either class of medication. Logistic regression was used to adjust the association of season of the year and flare of disease for potential confounding variables. Results: There was no association between season of the year and flare of Crohn’s disease (P ⴝ 0.66). Season of the year was only weakly associated with flares of ulcerative colitis (P ⴝ 0.02). Compared with winter, spring had very slightly higher rates of flares (OR ⴝ 1.13, 95% CI: 1.05– 1.23). We did not observe seasonal patterns within individual patients experiencing multiple flares (P > 0.05 for Crohn’s disease and ulcerative colitis). Conclusions: Although we observed a slight increase in exacerbations of ulcerative colitis in the spring, in general, these data do not support an association between season of the year and flares.
rohn’s disease and ulcerative colitis are collectively referred to as inflammatory bowel disease. Crohn’s disease and ulcerative colitis are both chronic relapsing diseases characterized by alternating periods of remission and active disease. Despite extensive research, no medication has been identified that can completely prevent relapse of inflammatory bowel disease. Because of this, it is critically important to gain a better understanding of the factors that contribute to relapse of these diseases.
C
In general, inflammatory bowel disease is believed to result from an imbalance between proinflammatory and anti-inflammatory forces. These biologic processes likely occur in response to a combination of genetic and environmental factors. Unfortunately, the environmental factors that contribute to the course of inflammatory bowel disease have not been well defined. Prior research has suggested that there may be seasonal variation in the natural history of inflammatory bowel disease, particularly among patients with ulcerative colitis, although these studies have yielded conflicting results.1–9 Likewise, patients anecdotally report that they tend to have disease relapses at similar times during the year. The purpose of this study was to determine whether, within a cohort of inflammatory bowel disease patients, relapse of inflammatory bowel disease follows a seasonal pattern. In addition, we tested the hypothesis that individual patients tend to repeatedly flare during the same time of each year.
Patients and Methods Patients We performed a retrospective cohort study among patients with inflammatory bowel disease identified from the General Practice Research Database. This study utilized data from the General Practice Research Database (GPRD) from January 1988 to October 1997. GPRD is composed of computerized medical records of approximately 2000 general practitioners (GPs) from the United Kingdom.10 Established in 1988, the system currently includes data on approximately 8,000,000 patients in total, representing 6% of the United Kingdom population.11 The age and sex distribution of the population served by GPRD physicians is similar to those of the general United Kingdom population.11 Abbreviations used in this paper: GP, general practitioner; GPRD, General Practice Research Database. © 2004 by the American Gastroenterological Association 0016-5085/04/$30.00 doi:10.1053/j.gastro.2003.12.003
666
LEWIS ET AL.
Within the United Kingdom, approximately 98% of the population are registered with a GP.10 This GP is responsible for almost the entirety of the patient’s medical care. Within GPRD, each GP undergoes formal training in the protocol of data entry. Each medical practice must demonstrate competency at entering data into the electronic database before their data are considered “up to standard.” Subsequently, each practice is subject to monthly audits to ensure that the data quality remains up to standard. Data recorded in the electronic record include demographic information, prescription information, clinical events and diagnoses, preventive care, hospital admissions, cause of death, and free text. As well, significant diagnoses occurring prior to the initiation of the electronic medical record are recorded retrospectively. Diagnoses are recorded using OXMIS (Oxford Medical Indexing System) codes, which can be cross-referenced to Read codes.10,12 Prescribed medications are recorded using codes issued by the Prescription Pricing Authority of the National Health Service.10,12 Several studies have shown that the clinical information in the computer record is sufficiently accurate for use in epidemiologic studies, including studies of inflammatory bowel disease.13–16
Study Cohorts All patients with a coded diagnosis of ulcerative colitis or Crohn’s disease (including regional enteritis, regional ileitis, or regional colitis) were potentially eligible for inclusion in the study (n ⫽ 22,611). Patients diagnosed only as “inflammatory bowel disease not otherwise specified” or with codes for both Crohn’s disease and ulcerative colitis were excluded (n ⫽ 2438) because it was not possible to know which diagnosis the patient carried and because we have previously demonstrated lower reliability of these diagnostic codes for inflammatory bowel disease.16 We also excluded patients with less than 1 year of follow-up time from registration, up to standard, and their first diagnosis with inflammatory bowel disease (n ⫽ 4627). Patients with rheumatoid arthritis, asthma, and chronic obstructive airway disease (n ⫽ 2059) were excluded from the cohort because these are common indications for steroid therapy and could reduce the predictive value of our algorithm for identifying flare of disease (described below). Likewise, patients who had evidence of a single flare but it occurred on the date of a diagnosis of a potential noninflammatory bowel disease indication for steroids, such as allergic reactions, were also excluded (n ⫽ 294). Patients who had no flares of inflammatory bowel disease identified during follow-up were also excluded (n ⫽ 8833).
Methods Identification of Inflammatory Bowel Disease Flare and Remission Period Follow-up time for this study began with the first prescription for corticosteroids, 5-ASA, or azathioprine/6-mercaptopurine that occurred at least 1 year after the letter of registration with the general practitioner, data up to standard,
GASTROENTEROLOGY Vol. 126, No. 3
and the first diagnosis with inflammatory bowel disease. For each patient, all follow-up time was categorized as representing a period of remission, flare, or unknown. Remission periods were identified when a patient who had previously received corticosteroids, 5-ASA medications, or azathioprine/6-mercaptopurine completed a period of 4 months with no prescriptions for corticosteroids or 5-ASA compounds. A flare of inflammatory bowel disease was identified by the first prescription for a 5-ASA medication or corticosteroid following a remission period. The period between the onset of a flare and the beginning of remission time was considered unknown in that the data in GPRD do not allow the investigator to know with certainty when a patient enters remission. The last 14 days prior to a flare were excluded from the remission time. Flares that occurred on the same date as a diagnosis for a potentially noninflammatory bowel disease indication for steroids (e.g., allergic reaction) were excluded.
Exposure Variables The primary exposure variable for this study was season of the year. We arbitrarily defined the seasons as follows: winter included December, January, February; spring included March, April, May; summer included June, July, August; autumn included September, October, November.
Reliability of the Algorithm for Identifying a Flare and Remission Periods From the GPRD data, it is often difficult to determine exactly when a patient completes a course of steroids. As such, we completed a survey of GPs caring for 150 patients with inflammatory bowel disease who received a new prescription for corticosteroids after a minimum of 4 months without a prior prescription for corticosteroids. The physician was asked whether this new prescription was for a new flare of inflammatory bowel disease. The GP was also queried as to whether the patient was still receiving corticosteroids 6 months after the original prescription.
Statistical Analyses All analyses were performed separately for patients with Crohn’s disease and ulcerative colitis. Continuous variables are reported as means with standard deviations. Categorical variables are reported as counts and percentages. Continuous variables were compared with the signed rank test.
Relationship of Season With Risk of Flare To test the association of season of year to risk of flare of IBD, we used logistic regression analysis. Month of the year was the independent variable, and flare was the dependent variable. Each month was coded as representing the corresponding season. To be coded as remission, there needed to be evidence of remission during the entire month. Months coded as unknown for flare or remission were excluded. The logistic regression analysis used a robust estimate of the variance clustering on the individual patient to account for lack of independence of the observations within a single subject.
March 2004
These analyses were adjusted for sex and calendar year of the observation. We sought to adjust the logistic regression analysis for the potentially confounding effects of use of prescription nonsteroidal anti-inflammatory drugs (NSAIDs), aspirin, azathioprine/6-mercaptopurine, and antibiotics. Because use of these drugs varies with time, we used a case-crossover design, matching each patient to himself.17 This also served to minimize confounding from additional, unmeasured covariates. For this analysis, we sampled 1 date from each remission period and selected the first date of all flares. We used conditional logistic regression matched on the subject with flare of disease as the dependent variable and season of the year as the independent variable. Potential confounder variables included use of prescription NSAIDs, aspirin, azathioprine/6-mercaptopurine, and antibiotics. These were all categorized as use within the preceding 60 days. All potential confounders were tested individually and then in a multivariate model for evidence of confounding of the association between season of the year and flare of disease. To evaluate whether the results of the study were robust to the specification of the algorithm defining flare, remission, and “unknown” time, we conducted 2 sensitivity analyses. We first examined the identity of the drugs used to define flare. Here, we repeated our analyses after defining flare as occurring only with a prescription for 5-ASA or only with a prescription for steroids. The second sensitivity analysis evaluated whether defining remission as occurring 4 months after a prescription (vs. a longer period) was important in driving the study results. Here, we use Monte Carlo simulation. Each subject had his date of flare adjusted forward (beyond 4 months) by a random number sampled from a half-normal distribution with a mean of zero and 4 standard deviations equal to 90 days. A halfnormal distribution was used because the date of the new prescription for steroids or 5-ASA medications could only come after the start of the flare, not before. The logistic regression model was then run using these data to calculate odds ratios for the association between season of the year and flare of disease. This procedure was repeated 500 times, with the mean and range of the odds ratios reported.
Relationship of Repeated Flares During the Same Season Within Individuals To determine whether individual patients tend to have flares during the same season of the year, we examined subgroups of the cohorts who experienced more than 1 flare. We required a minimum of 2 years of follow-up time for those with 2 flares, 3 years for those with 3 flares, and 4 years for those with 4 flares. We then determined the maximum number of flares that occurred in the same season for each individual. The distribution of the maximal number of flares in any one season was then compared with the expected multinomial distribution, assuming that the season of flare of disease was independent, using the goodness-of-fit test. To further assess this, we used generalized estimating equations (GEE) to ex-
SEASONALITY OF IBD
667
Table 1. Characteristics of the Study Cohorts Crohn’s disease (n ⫽ 1587) Gender (% female) Age at start of follow-up (yr), mean (SD) Duration of follow-up (yr), median (IQR) Time to first flare (days), median (IQR) Number of flares during follow-up, median (IQR) Months of remission off medications, median (IQR) Months receiving 5-ASA or steroids, median (IQR) Months excluded for definition of remissiona or other criteria,b median (IQR)
61.6% 41.4 (16.6)
Ulcerative colitis (n ⫽ 2773) 50.5% 47.7 (15.8)
3.8 (2.0–5.2)
4.0 (2.4–5.4)
225 (17–594)
244 (53–620)
1 (1–2)
1 (1–2)
6 (0–20)
7 (1–22)
16 (4–35)
17 (4–37)
7 (5–11)
7 (5–13)
IQR, interquartile range. aA period of 4 months with no prescriptions for corticosteroids or 5-ASA compounds was required for a definition of remission. bMonths were excluded if a new prescription for steroids or 5-ASA occurred on a date with a diagnosis of a potential non-IBD indication for steroids, e.g., allergic reaction or if the patient was deemed to be in remission for only part of the month.
amine the auto regressive correlational structure of the logistic regression model. All analyses were conducted using 2-sided tests. Statistical significance was set at a P value of 0.05. All analyses were performed using SAS version 8.1 (SAS Institute, Cary, NC) and Stata version 7.0 (Stata Corp, College Station, TX).
Results The final study cohort included 1587 patients with Crohn’s disease (62% female; mean age at start of follow-up, 41 ⫾ 17 years) and 2773 patients with ulcerative colitis (50% female; mean age at start of followup, 48 ⫾ 16 years) (Table 1). The median duration from the last prescription for inflammatory bowel disease therapies to the start of a flare was 247 days (interquartile range, 175 to 442 days) for Crohn’s disease and 243 days (interquartile range, 172 to 440 days) for ulcerative colitis. Identification of Flare of Inflammatory Bowel Disease Completed surveys with usable data were received from GPs on 136 of the 150 patients (91%). The 136 patients (79 female and 57 male) included 59 with Crohn’s disease, 68 with ulcerative colitis, 5 with inflammatory bowel disease not otherwise specified, and 4 in whom the GP reported that the patient did not have inflammatory bowel disease. The mean age on the index
668
LEWIS ET AL.
GASTROENTEROLOGY Vol. 126, No. 3
Table 2. Odds Ratio for the Association Between Season of the Year and Flare of Disease Crohn’s disease
Winter Spring Summer Autumn
Ulcerative colitis
Model 1 OR (95% CI)
Model 2 OR (95% CI)
Model 1 OR (95% CI)
Model 2 OR (95% CI)
Reference 1.06 (0.95–1.18) 1.07 (0.96–1.19) 1.05 (0.94–1.18)
Reference 1.02 (0.86–1.21) 1.04 (0.87–1.23) 1.07 (0.90–1.27)
Reference 1.13 (1.05–1.23) 1.07 (0.99–1.16) 1.08 (0.99–1.17)
Reference 1.14 (1.01–1.28) 1.07 (0.94–1.21) 1.06 (0.94–1.20)
NOTE. Model 1: Logistic regression model adjusted for calendar year and sex. Model 2: Conditional logistic regression model adjusted for calendar year, aspirin, NSAID, azathioprine/6-mercaptopurine, and antibiotic use.
date was 47 ⫾ 20 years. After excluding patients with a previous diagnosis of rheumatoid arthritis, chronic obstructive airway disease, or asthma, the positive predictive value of a new prescription for steroids to identify an acute flare of inflammatory bowel disease was 82 of 96 (85%). We used each patient’s multiple episodes of drug use to examine whether a 4-month window was sufficient to identify a patient no longer receiving steroids. Using this algorithm, the positive predictive value of no longer taking steroids 4 months after the last prescription was 29 of 32 (91%). To evaluate whether prescriptions identifying a flare might be routine refills for patients in remission, we examined the timing of prescriptions before and after a flare. The first interval was the median time from the last prescription for steroids or 5-ASA therapy until a new flare was identified. The second interval was the median time from the prescription identifying a flare to the next prescription for steroids or 5-ASA. The interval between prescriptions was markedly shorter after the flare for both Crohn’s disease (median, 183 days [interquartile range, 143 to 312 days] before a flare vs. 71 days [interquartile range, 33 to 127 days] after a flare, P ⬍ 0.0001) and ulcerative colitis (median, 180 days [interquartile range,
Figure 1. Seasonality of flares within the cohort of Crohn’s disease patients.
141 to 297 days] before a flare vs. 84 days [interquartile range, 36 to 140 days] after a flare, P ⬍ 0.0001). This suggests that prescriptions identifying a flare are unlikely to be routine refills for patients in remission. Association of Season of the Year and Flare Within the Cohort of Inflammatory Bowel Disease Patients In the logistic regression model adjusting only for calendar year and sex, there was no association between season of the year and flare of Crohn’s disease (P ⫽ 0.66 for overall comparison of all seasons) (Table 2 and Figure 1). For ulcerative colitis, the model showed a very weak but statistically significant association between season of the year and flare of disease (P ⫽ 0.02) (Figure 2). Using winter as the reference category, only spring had a higher rate of flare of disease (OR ⫽ 1.13, 95% CI: 1.05–1.23). The conditional logistic regression models adjusting for calendar year, NSAID, aspirin, azathioprine, and antibiotic exposure produced nearly identical results (Table 2). Sensitivity analyses were performed in which flares were identified only by a prescription for a 5-ASA product or only a prescription for steroids using the fully
Figure 2. Seasonality of flares within the cohort of ulcerative colitis patients. Using winter as the reference category, only spring had a slightly higher rate of flare of disease (OR ⫽ 1.13, 95% CI: 1.05– 1.23).
March 2004
SEASONALITY OF IBD
Table 3. Sensitivity Analyses Examining Different Definition of Flare of Crohn’s Disease 5-ASA or steroid OR (95% CI) Winter Spring Summer Autumn
5-ASA only OR (95% CI)
Table 4. Sensitivity Analyses Examining Different Definition of Flare of Ulcerative Colitis
Steroid only OR (95% CI)
Reference Reference Reference 1.02 (0.86–1.21) 0.99 (0.79–1.23) 1.06 (0.83–1.35) 1.04 (0.87–1.23) 1.09 (0.88–1.36) 0.96 (0.75–1.21) 1.07 (0.90–1.27) 1.03 (0.83–1.29) 1.20 (0.94–1.53)
adjusted conditional logistic regression models. In all cases, although the confidence intervals are wider because of reduced number of observations, similar results were obtained in each model (Tables 3 and 4). Likewise, sensitivity analyses were conducted in which we excluded patients with multiple flares, at least one of which occurred on the same date as a diagnosis for a potentially noninflammatory bowel disease indication for steroids (n ⫽ 577 for Crohn’s disease and n ⫽ 989 for ulcerative colitis). These analyses again resulted in nearly identical results (data not shown). A second set of sensitivity analyses was conducted in which we required longer periods without a prescription for steroids or 5-ASA products to identify new flares of disease in our model adjusted for calendar year and sex. Increasing this time period from 4 months to 6 months or 8 months had a minimal effect on the results. The greatest absolute change in the estimated odds ratio was 0.11 (risk of flare of ulcerative colitis in the spring increased from 1.13 to 1.24 in the 8-month model). In no case did any nonstatistically significant results become statistically significant. In another sensitivity analysis, we required a minimum of 2 prescriptions for steroids or 5-ASA products within a 60-day window to identify a new flare. Again, the results were nearly identical to the primary analyses. The largest absolute difference in the odds ratio from the primary analyses was 0.08 (risk of flare of Crohn’s disease in the spring increased from 1.06 to 1.14). Additional sensitivity analyses were performed using a Monte Carlo simulation to examine the potential effect of misclassification of the dates of flares. The mean odds ratios based on 500 simulations, in which random adjustment forward was made to the date of flare according to the half-normal distribution, were very close to our original estimates (Table 5). In addition, the distribution of the odds ratios from the Monte Carlo simulation suggested that misclassification of the date of a flare was unlikely to affect substantially our conclusions. The largest absolute difference between the odds ratio from our primary analysis and the upper bound of the range of odds ratios from the simulation was only 0.12 (Crohn’s disease flare in spring).
669
5-ASA or steroid OR (95% CI) Winter Spring Summer Autumn
5-ASA only OR (95% CI)
Steroid only OR (95% CI)
Reference Reference Reference 1.14 (1.01–1.28) 1.14 (0.99–1.30) 1.09 (0.88–1.34) 1.07 (0.94–1.21) 1.05 (0.92–1.20) 0.99 (0.80–1.22) 1.06 (0.94–1.20) 1.04 (0.91–1.19) 1.10 (0.89–1.36)
Association of Season of the Year and Flare Within Individual Inflammatory Bowel Disease Patients To determine whether individual patients tend to have flares during the same season of the year, we examined subgroups of the cohorts who experienced more than 1 flare. Among those patients with 2 flares, 3 flares, and 4 flares, the concurrence of multiple flares within a single season was no greater than expected for either Crohn’s disease or ulcerative colitis (P ⬎ 0.05 for all comparisons) (Table 6 and Figures 3 and 4). Of note, although this analysis approached statistical significance in patients with 2 flares, this was due to these flares occurring less commonly than expected in the same season. To further assess the relationship between season of flares within individual subjects, we fit a version of our GEE logistic regression model using autoregressive correlational structure (AR4). The model parameters and associated P values were not significantly impacted by the change in correlational structure applied by using the autoregressive structure. The correlations of importance that are estimated in the model are the correlations between flare status for 2 seasons separated by 1 season, 2 seasons, 3 seasons, and 1 year. If individual patients tended to flare during the same season, we would expect to see higher correlation coefficients in the fourth season after the initial flare. In fact, the correlation coefficients were comparable among the second, third, and fourth Table 5. Comparison of Results From the Primary Analyses to the Monte Carlo Simulation
Season Crohn’s disease Spring Summer Autumn Ulcerative colitis Spring Summer Autumn
Odds ratio from primary analyses
Mean odds ratio from Monte Carlo simulation
Range of odds ratios from Monte Carlo simulation
1.06 1.07 1.05
1.09 1.08 1.04
1.00–1.18 0.99–1.16 0.95–1.14
1.13 1.07 1.08
1.08 1.05 1.02
1.02–1.15 1.01–1.10 0.96–1.08
670
LEWIS ET AL.
GASTROENTEROLOGY Vol. 126, No. 3
Table 6. Periodicity of Flares Within Individual Subjects Number of flares per subject
Crohn’s disease patients (n)
Crohn’s disease P value
Ulcerative colitis patients (n)
Ulcerative colitis P value
2 3 4
266 120 74
0.06 0.23 0.25
506 242 130
0.09 0.64 0.48
seasons following the flare (0.05 to 0.06 for ulcerative colitis and 0.03 for all 3 seasons for Crohn’s disease). These results provide further support that individual patients do not tend to flare during the same season.
Discussion There are several reasons to hypothesize that inflammatory bowel disease may have a seasonal pattern. Environmental exposures vary with seasons of the year. These environmental exposures could influence the natural history of inflammatory bowel disease. For example, use of medications that may influence the natural history of inflammatory bowel disease, such as NSAIDs,18 –20 may vary across seasons. Perhaps more importantly, exposure to antigens such as viral infections vary according to season of the year.21,22 Nearly 70 years ago, Banks and Bargen suggested that upper respiratory tract infection may predispose patients to exacerbation of otherwise quiescent ulcerative colitis.23 They noted these infections in 60% of patients who relapsed. Similar results have been reported by a few other investigators,9,24 whereas others have failed to identify an association between upper respiratory tract infection and relapse of inflam-
Figure 3. Time from first flare to second flare among patients with Crohn’s disease who experienced more than 1 flare. Each bar represents a period of 100 days. The first bar represents the period from day 101 to 200. There is no bar for the period from day 1 to 100 because our algorithm required a minimum of 122 days between prescriptions to identify a new flare.
Figure 4. Time from first flare to second flare among patients with ulcerative colitis who experienced more than 1 flare. Each bar represents a period of 100 days. The first bar represents the period from day 101 to 200. There is no bar for the period from day 1 to 100 because our algorithm required a minimum of 122 days between prescriptions to identify a new flare.
matory bowel disease.4 These studies have generally been limited by small sample sizes and in some cases by inability to determine exact temporal relationships between upper respiratory tract infection and relapse of inflammatory bowel disease. Nonetheless, they provide reason to believe that relapse of inflammatory bowel disease could be seasonal. Previous research has resulted in conflicting opinions as to whether inflammatory bowel disease is more likely to flare during selected periods of the year (Table 7).1–9 Two prior studies demonstrated increased rates of ulcerative colitis flares in the spring,6,9 whereas other studies have suggested increased rates of flares in the autumn or winter.1,2,4 Still other studies have suggested that there is no pattern of flare among patients with ulcerative colitis.3,7 The results for Crohn’s disease have been equally as inconsistent.1,3,5,7,9 In addition, patients frequently report that their disease tends to be more active during certain periods of the year. Our study does not support an association between season of the year and flares of disease either within the cohort or within individual patients with inflammatory bowel disease. Although there was a statistically significant increase in the risk of flare of ulcerative colitis in the spring, the magnitude of this association was extremely small (OR ⫽ 1.14). Given the numerous statistical tests performed, it is likely that this was a chance finding (i.e., type 1 error). Furthermore, if there is an association between season of the year and flare of disease, we would have expected to also see evidence of this association among individual patients with multiple
March 2004
SEASONALITY OF IBD
671
Table 7. Seasonal Pattern of IBD Exacerbation Author and year 1
Myszor and Calam, 1984 Sellu, 19862 Don and Goldacre, 19843 Riley et al., 19904 Zeng and Anderson, 19965 Tysk and Jarnerof, 19936 Sonnenberg et al., 19947 Sonnenberg and Wassermann, 19918 Kangro et al., 19909a aCD
Location
CD peak
UC peak
United Kingdom United Kingdom United Kingdom United Kingdom Canada Sweden USA USA
No pattern Not reported No pattern Not reported September–February Not Reported No pattern Pattern similar to all other diagnoses
September–February September–February No pattern August–January Not reported Spring and Autumn No pattern Pattern similar to all other diagnoses
United Kingdom
Winter, April–May, and July–August
Winter, April–May, and July–August
and UC data were combined in this study.
flares. In contrast, when we limited our analyses to patients with multiple flares, we saw no evidence of a seasonal pattern of either Crohn’s disease or ulcerative colitis. Thus, we feel that these data argue against a seasonal pattern to the natural history of inflammatory bowel disease. Our study has several features that make it unique compared with the prior studies. Most notable is that a large cohort of patients was followed for an extended period of time, such that we could account for the effect of multiple relapses among an individual patient. Also, to our knowledge, this is the first study to examine the periodicity of relapse within individual subjects. Another unique feature of our study was the ability to adjust for changes in medication exposures over the course of multiple flares and remissions. Medications such as antibiotics and NSAIDs could potentially influence relapse rates of inflammatory bowel disease. Failure to adjust for these medications could bias results of studies addressing seasonality of relapses. Because the GPRD contains data on medications prescribed, we were able to account for the variable use of medications across the course of the study. Our data were from the records of primary care physicians, not gastroenterologists. As such, this population likely contains more patients with mild disease than is seen in referral centers. Importantly, we have previously validated the diagnosis of inflammatory bowel disease within the GPRD and demonstrated that the prevalence, anatomic distribution of disease, and prescribing patterns for inflammatory bowel disease patients within GPRD are consistent with that observed in other populations.16 Furthermore, in this study, we performed additional validation studies to examine the reliability of using medication prescriptions to identify exacerbations of inflammatory bowel disease. Our results demonstrated relatively high predictive values for steroid prescriptions to identify flares of inflammatory bowel disease. Importantly, however, we used new 5-ASA prescriptions to
identify flares as well. We elected not to validate the new 5-ASA prescriptions as a marker of disease relapse, in part because of financial constraints. Given that these medications are much more specific for inflammatory bowel disease than are steroids, the predictive value of a new 5-ASA prescription to identify a disease relapse should be even higher than that of a new steroid prescription. Any nondifferential misclassification of remission periods as flares would be expected to bias our results toward the null. However, and importantly, when we reanalyzed our data using only steroids or only 5-ASA medications to identify flares of disease, we obtained nearly identical results. As such, we feel confident that our algorithm is not significantly biased by including both 5-ASA and steroids as a marker of relapse. It is possible that a small proportion of flares may have gone undocumented, either because the symptoms were so mild that the patients did not seek treatment or if the flare was treated by other physicians. Within our validation study, we documented that more than 83% of the patients with Crohn’s disease and 80% of those with ulcerative colitis were receiving care from a gastroenterologist as well as from their GP (data not shown). However, several features of our study make this unlikely to have significantly biased our results. First, if flares went completely unrecorded because they were treated exclusively by the gastroenterologist, this pattern would need to occur preferentially in the seasons with the highest rates of flare to mask a true association. There is no reason to believe that flares in one season of the year would be more likely to be treated exclusively by a gastroenterologist than flares in other seasons of the year. Second, within the United Kingdom, specialists usually prescribe a small quantity of medication, and refills are written by the GP. All of our patients had received prescriptions for inflammatory bowel disease medications from their GP as a requirement for inclusion. Thus, these GPs are accustomed to writing for steroids and 5-ASA products. In addition, if some flares were initially treated
672
LEWIS ET AL.
by the gastroenterologist, it is expected that the GP would write the continuing prescriptions for these medications. Although this may slightly change the estimated date of the flare, our Monte Carlo simulation demonstrates that it would be unlikely to substantially change the results. Thus, we believe that misclassification related to identification and dating the start of flares is unlikely to have significantly affected our results. It is possible that a few of the prescriptions that we used to identify a flare could actually have been a refill for a patient in remission. Importantly, when we used even more stringent definitions of a flare (e.g., an 8-month interval without prescriptions for steroids or 5-ASA products), we obtained nearly identical results. Thus, this form of misclassification is unlikely to have significantly influenced our results. Another potential limitation of our data is the possibility of incomplete adjusting for confounders. Although we were able to adjust for many factors potentially associated with relapse of disease, there may be other factors not accounted for in our analyses. Our case crossover analyses matching each patient to herself or himself as a control should minimize the effect of potential confounders that are unlikely to vary across time, such as smoking habits. We elected to exclude patients with rheumatoid arthritis, asthma, and chronic obstructive airway disease from our analyses so as to improve the specificity of our algorithm to identify flares of inflammatory bowel disease. It is possible that the subgroup of patients with both inflammatory bowel disease and reactive airway disease are uniquely sensitive to seasonal variations. However, when we include those subjects in the analyses, we obtain very similar results for both within cohort and within subject comparisons (data not shown). Likewise, we excluded patients with less than 1 year of follow-up and started our follow-up time at least 1 year after the first diagnosis with inflammatory bowel disease. This was done to avoid potential bias associated with newly diagnosed patients or from inclusion of patients who received only short-term care from the general practitioner. Importantly, when we repeated the analyses including these patients, the results were again nearly identical (data not shown). As such, it seems unlikely that exclusion of either of these groups of patients has biased our results. We excluded patients who were chronically maintained on steroids and/or 5-ASA medications. This was necessary to be able to identify a flare of disease. It is possible, albeit unlikely, that such patients would have a different seasonal pattern of relapse than patients who
GASTROENTEROLOGY Vol. 126, No. 3
discontinue these medications. Unfortunately, we are unable to address this question within this cohort. In conclusion, our data do not support a seasonal pattern to the natural history of inflammatory bowel disease. We found little evidence of seasonality of relapse within a cohort of patients with inflammatory bowel disease or within individual inflammatory bowel disease patients. Although our data do not exclude the possibility that environmental exposures such as upper respiratory tract infections or their treatments may trigger a relapse of inflammatory bowel disease, it seems less likely that environmental exposures that vary with season of the year explain a large proportion of the variance in the natural history of this disease.
References 1. Myszor M, Calam J. Seasonality of ulcerative colitis. Lancet 1984;2:522–523. 2. Sellu DP. Seasonal variation in onset of exacerbations of ulcerative proctocolitis. J Royal Col Surg Edinburgh 1986;31:158 – 160. 3. Don BA, Goldacre MJ. Absence of seasonality in emergency hospital admissions for inflammatory bowel disease. Lancet 1984;2:1156 –1157. 4. Riley SA, Mani V, Goodman MJ, Lucas S. Why do patients with ulcerative colitis relapse? Gut 1990;31:179 –183. 5. Zeng L, Anderson FH. Seasonal change in the exacerbations of Crohn’s disease. Scand J Gastroenterol 1996;31:79 – 82. 6. Tysk C, Jarnerot G. Seasonal variation in exacerbations of ulcerative colitis. Scand J Gastroenterol 1993;28:95–96. 7. Sonnenberg A, Jacobsen SJ, Wasserman IH. Periodicity of hospital admissions for inflammatory bowel disease. Am J Gastroenterol 1994;89:847– 851. 8. Sonnenberg A, Wasserman IH. Epidemiology of inflammatory bowel disease among U.S. military veterans. Gastroenterology 1991;101:122–130. 9. Kangro HO, Chong SK, Hardiman A, Heath RB, Walker-Smith JA. A prospective study of viral and mycoplasma infections in chronic inflammatory bowel disease. Gastroenterology 1990;98:549 – 553. 10. Lis Y, Mann RD. The VAMP Research multi-purpose database in the U.K. J Clin Epidemiol 1995;48:431– 443. 11. Anonymous. The General Practice Research Database: Information for Researchers. Office for National Statistics, 1994. 12. Garcia Rodriguez LA, Perez-Gutthann S. Use of the UK General Practice Research Database for pharmacoepidemiology. Br J Clin Pharmacol 1998;45:419 – 425. 13. Jick H, Jick SS, Derby LE. Validation of information recorded on general practitioner based computerised data resource in the United Kingdom. BMJ 1991;302:766 –768. 14. Jick H, Terris B, Derby L, Jick S. Further validation of information recorded on a general practitioner based computerized data resource in the United Kingdom. Pharmacoepidemiol Drug Safety 1992;1:347–349. 15. Garcia Rodriguez L, Perez-Gutthann S, Jick S. The UK General Practice Research Database. In: Strom B, ed. Pharmacoepidemiology. 3rd ed. New York: John Wiley & Sons, 2000:375–385. 16. Lewis JD, Brensinger C, Bilker WB, Strom BL. Validity and completeness of the General Practice Research Database for studies of inflammatory bowel disease. Pharmacoepidemiol Drug Safety 2002;11:211–218. 17. Maclure M. The case-crossover design: a method for studying
March 2004
18.
19.
20.
21.
transient effects on the risk of acute events. Am J Epidemiol 1991;133:144 –153. Gibson GR, Whitacre EB, Ricotti CA. Colitis induced by nonsteroidal anti-inflammatory drugs. Report of four cases and review of the literature. Arch Intern Med 1992;152:625– 632. Evans JM, McMahon AD, Murray FE, McDevitt DG, MacDonald TM. Non-steroidal anti-inflammatory drugs are associated with emergency admission to hospital for colitis due to inflammatory bowel disease. Gut 1997;40:619 – 622. Felder JB, Korelitz BI, Rajapakse R, Schwarz S, Horatagis AP, Gleim G. Effects of nonsteroidal antiinflammatory drugs on inflammatory bowel disease: a case-control study. Am J Gastroenterol 2000;95:1949 –1954. Dowell SF. Seasonal variation in host susceptibility and cycles of certain infectious diseases. Emerg Infect Dis 2001;7:369 –374.
SEASONALITY OF IBD
673
22. Noah ND. Cyclical patterns and predictability in infection. Epidemiol Infect 1989;102:175–190. 23. Banks B, Bargen J. Relapses in chronic ulcerative colitis— causes and prevention. Arch Intern Med 1934;53:131–139. 24. Mee AS, Jewell DP. Factors inducing relapse in inflammatory bowel disease. BMJ 1978;2:801– 802.
Received June 2, 2003. Accepted December 4, 2003. Address requests for reprints to: James D. Lewis, M.D., MSCE, University of Pennsylvania, Center for Clinical Epidemiology and Biostatistics, 7th Floor Blockley Hall, 423 Guardian Drive, Philadelphia, Pennsylvania 19104-6021. e-mail: [email protected]; fax: (215) 573-5325. Supported by National Institutes of Health grant K08-DK02589.
Seasonal Variation in Flares of Inflammatory Bowel Disease JAMES D. LEWIS,*,‡,§,㛳 FATEN N. ABERRA,‡ GARY R. LICHTENSTEIN,‡ WARREN B. BILKER,*,§,㛳 COLLEEN BRENSINGER,*,§ and BRIAN L. STROM*,‡,§,㛳,¶ *Center for Clinical Epidemiology and Biostatistics; ‡Department of Medicine; §Department of Biostatistics and Epidemiology; 㛳University of Pennsylvania Centers for Education and Research on Therapeutics; and ¶Department of Pharmacology, University of Pennsylvania School of Medicine, Philadelphia, Pennsylvania
Background & Aims: Previous research has yielded conflicting data as to whether the natural history of inflammatory bowel disease follows a seasonal pattern. The purpose of this study was to determine whether relapse of inflammatory bowel disease follows a seasonal pattern either across a cohort of patients or within individual patients. Methods: We used 1988 to 1997 data from the General Practice Research Database to conduct a retrospective cohort study of 1587 patients with Crohn’s disease (mean age at start of follow-up, 41 ⴞ 17 years) and 2773 patients with ulcerative colitis (mean age at start of follow-up, 48 ⴞ 16 years). Flares of disease were identified by receipt of a new prescription for either corticosteroids or 5-ASA medications following an interval of at least 4 months without prescriptions for either class of medication. Logistic regression was used to adjust the association of season of the year and flare of disease for potential confounding variables. Results: There was no association between season of the year and flare of Crohn’s disease (P ⴝ 0.66). Season of the year was only weakly associated with flares of ulcerative colitis (P ⴝ 0.02). Compared with winter, spring had very slightly higher rates of flares (OR ⴝ 1.13, 95% CI: 1.05– 1.23). We did not observe seasonal patterns within individual patients experiencing multiple flares (P > 0.05 for Crohn’s disease and ulcerative colitis). Conclusions: Although we observed a slight increase in exacerbations of ulcerative colitis in the spring, in general, these data do not support an association between season of the year and flares.
rohn’s disease and ulcerative colitis are collectively referred to as inflammatory bowel disease. Crohn’s disease and ulcerative colitis are both chronic relapsing diseases characterized by alternating periods of remission and active disease. Despite extensive research, no medication has been identified that can completely prevent relapse of inflammatory bowel disease. Because of this, it is critically important to gain a better understanding of the factors that contribute to relapse of these diseases.
C
In general, inflammatory bowel disease is believed to result from an imbalance between proinflammatory and anti-inflammatory forces. These biologic processes likely occur in response to a combination of genetic and environmental factors. Unfortunately, the environmental factors that contribute to the course of inflammatory bowel disease have not been well defined. Prior research has suggested that there may be seasonal variation in the natural history of inflammatory bowel disease, particularly among patients with ulcerative colitis, although these studies have yielded conflicting results.1–9 Likewise, patients anecdotally report that they tend to have disease relapses at similar times during the year. The purpose of this study was to determine whether, within a cohort of inflammatory bowel disease patients, relapse of inflammatory bowel disease follows a seasonal pattern. In addition, we tested the hypothesis that individual patients tend to repeatedly flare during the same time of each year.
Patients and Methods Patients We performed a retrospective cohort study among patients with inflammatory bowel disease identified from the General Practice Research Database. This study utilized data from the General Practice Research Database (GPRD) from January 1988 to October 1997. GPRD is composed of computerized medical records of approximately 2000 general practitioners (GPs) from the United Kingdom.10 Established in 1988, the system currently includes data on approximately 8,000,000 patients in total, representing 6% of the United Kingdom population.11 The age and sex distribution of the population served by GPRD physicians is similar to those of the general United Kingdom population.11 Abbreviations used in this paper: GP, general practitioner; GPRD, General Practice Research Database. © 2004 by the American Gastroenterological Association 0016-5085/04/$30.00 doi:10.1053/j.gastro.2003.12.003
666
LEWIS ET AL.
Within the United Kingdom, approximately 98% of the population are registered with a GP.10 This GP is responsible for almost the entirety of the patient’s medical care. Within GPRD, each GP undergoes formal training in the protocol of data entry. Each medical practice must demonstrate competency at entering data into the electronic database before their data are considered “up to standard.” Subsequently, each practice is subject to monthly audits to ensure that the data quality remains up to standard. Data recorded in the electronic record include demographic information, prescription information, clinical events and diagnoses, preventive care, hospital admissions, cause of death, and free text. As well, significant diagnoses occurring prior to the initiation of the electronic medical record are recorded retrospectively. Diagnoses are recorded using OXMIS (Oxford Medical Indexing System) codes, which can be cross-referenced to Read codes.10,12 Prescribed medications are recorded using codes issued by the Prescription Pricing Authority of the National Health Service.10,12 Several studies have shown that the clinical information in the computer record is sufficiently accurate for use in epidemiologic studies, including studies of inflammatory bowel disease.13–16
Study Cohorts All patients with a coded diagnosis of ulcerative colitis or Crohn’s disease (including regional enteritis, regional ileitis, or regional colitis) were potentially eligible for inclusion in the study (n ⫽ 22,611). Patients diagnosed only as “inflammatory bowel disease not otherwise specified” or with codes for both Crohn’s disease and ulcerative colitis were excluded (n ⫽ 2438) because it was not possible to know which diagnosis the patient carried and because we have previously demonstrated lower reliability of these diagnostic codes for inflammatory bowel disease.16 We also excluded patients with less than 1 year of follow-up time from registration, up to standard, and their first diagnosis with inflammatory bowel disease (n ⫽ 4627). Patients with rheumatoid arthritis, asthma, and chronic obstructive airway disease (n ⫽ 2059) were excluded from the cohort because these are common indications for steroid therapy and could reduce the predictive value of our algorithm for identifying flare of disease (described below). Likewise, patients who had evidence of a single flare but it occurred on the date of a diagnosis of a potential noninflammatory bowel disease indication for steroids, such as allergic reactions, were also excluded (n ⫽ 294). Patients who had no flares of inflammatory bowel disease identified during follow-up were also excluded (n ⫽ 8833).
Methods Identification of Inflammatory Bowel Disease Flare and Remission Period Follow-up time for this study began with the first prescription for corticosteroids, 5-ASA, or azathioprine/6-mercaptopurine that occurred at least 1 year after the letter of registration with the general practitioner, data up to standard,
GASTROENTEROLOGY Vol. 126, No. 3
and the first diagnosis with inflammatory bowel disease. For each patient, all follow-up time was categorized as representing a period of remission, flare, or unknown. Remission periods were identified when a patient who had previously received corticosteroids, 5-ASA medications, or azathioprine/6-mercaptopurine completed a period of 4 months with no prescriptions for corticosteroids or 5-ASA compounds. A flare of inflammatory bowel disease was identified by the first prescription for a 5-ASA medication or corticosteroid following a remission period. The period between the onset of a flare and the beginning of remission time was considered unknown in that the data in GPRD do not allow the investigator to know with certainty when a patient enters remission. The last 14 days prior to a flare were excluded from the remission time. Flares that occurred on the same date as a diagnosis for a potentially noninflammatory bowel disease indication for steroids (e.g., allergic reaction) were excluded.
Exposure Variables The primary exposure variable for this study was season of the year. We arbitrarily defined the seasons as follows: winter included December, January, February; spring included March, April, May; summer included June, July, August; autumn included September, October, November.
Reliability of the Algorithm for Identifying a Flare and Remission Periods From the GPRD data, it is often difficult to determine exactly when a patient completes a course of steroids. As such, we completed a survey of GPs caring for 150 patients with inflammatory bowel disease who received a new prescription for corticosteroids after a minimum of 4 months without a prior prescription for corticosteroids. The physician was asked whether this new prescription was for a new flare of inflammatory bowel disease. The GP was also queried as to whether the patient was still receiving corticosteroids 6 months after the original prescription.
Statistical Analyses All analyses were performed separately for patients with Crohn’s disease and ulcerative colitis. Continuous variables are reported as means with standard deviations. Categorical variables are reported as counts and percentages. Continuous variables were compared with the signed rank test.
Relationship of Season With Risk of Flare To test the association of season of year to risk of flare of IBD, we used logistic regression analysis. Month of the year was the independent variable, and flare was the dependent variable. Each month was coded as representing the corresponding season. To be coded as remission, there needed to be evidence of remission during the entire month. Months coded as unknown for flare or remission were excluded. The logistic regression analysis used a robust estimate of the variance clustering on the individual patient to account for lack of independence of the observations within a single subject.
March 2004
These analyses were adjusted for sex and calendar year of the observation. We sought to adjust the logistic regression analysis for the potentially confounding effects of use of prescription nonsteroidal anti-inflammatory drugs (NSAIDs), aspirin, azathioprine/6-mercaptopurine, and antibiotics. Because use of these drugs varies with time, we used a case-crossover design, matching each patient to himself.17 This also served to minimize confounding from additional, unmeasured covariates. For this analysis, we sampled 1 date from each remission period and selected the first date of all flares. We used conditional logistic regression matched on the subject with flare of disease as the dependent variable and season of the year as the independent variable. Potential confounder variables included use of prescription NSAIDs, aspirin, azathioprine/6-mercaptopurine, and antibiotics. These were all categorized as use within the preceding 60 days. All potential confounders were tested individually and then in a multivariate model for evidence of confounding of the association between season of the year and flare of disease. To evaluate whether the results of the study were robust to the specification of the algorithm defining flare, remission, and “unknown” time, we conducted 2 sensitivity analyses. We first examined the identity of the drugs used to define flare. Here, we repeated our analyses after defining flare as occurring only with a prescription for 5-ASA or only with a prescription for steroids. The second sensitivity analysis evaluated whether defining remission as occurring 4 months after a prescription (vs. a longer period) was important in driving the study results. Here, we use Monte Carlo simulation. Each subject had his date of flare adjusted forward (beyond 4 months) by a random number sampled from a half-normal distribution with a mean of zero and 4 standard deviations equal to 90 days. A halfnormal distribution was used because the date of the new prescription for steroids or 5-ASA medications could only come after the start of the flare, not before. The logistic regression model was then run using these data to calculate odds ratios for the association between season of the year and flare of disease. This procedure was repeated 500 times, with the mean and range of the odds ratios reported.
Relationship of Repeated Flares During the Same Season Within Individuals To determine whether individual patients tend to have flares during the same season of the year, we examined subgroups of the cohorts who experienced more than 1 flare. We required a minimum of 2 years of follow-up time for those with 2 flares, 3 years for those with 3 flares, and 4 years for those with 4 flares. We then determined the maximum number of flares that occurred in the same season for each individual. The distribution of the maximal number of flares in any one season was then compared with the expected multinomial distribution, assuming that the season of flare of disease was independent, using the goodness-of-fit test. To further assess this, we used generalized estimating equations (GEE) to ex-
SEASONALITY OF IBD
667
Table 1. Characteristics of the Study Cohorts Crohn’s disease (n ⫽ 1587) Gender (% female) Age at start of follow-up (yr), mean (SD) Duration of follow-up (yr), median (IQR) Time to first flare (days), median (IQR) Number of flares during follow-up, median (IQR) Months of remission off medications, median (IQR) Months receiving 5-ASA or steroids, median (IQR) Months excluded for definition of remissiona or other criteria,b median (IQR)
61.6% 41.4 (16.6)
Ulcerative colitis (n ⫽ 2773) 50.5% 47.7 (15.8)
3.8 (2.0–5.2)
4.0 (2.4–5.4)
225 (17–594)
244 (53–620)
1 (1–2)
1 (1–2)
6 (0–20)
7 (1–22)
16 (4–35)
17 (4–37)
7 (5–11)
7 (5–13)
IQR, interquartile range. aA period of 4 months with no prescriptions for corticosteroids or 5-ASA compounds was required for a definition of remission. bMonths were excluded if a new prescription for steroids or 5-ASA occurred on a date with a diagnosis of a potential non-IBD indication for steroids, e.g., allergic reaction or if the patient was deemed to be in remission for only part of the month.
amine the auto regressive correlational structure of the logistic regression model. All analyses were conducted using 2-sided tests. Statistical significance was set at a P value of 0.05. All analyses were performed using SAS version 8.1 (SAS Institute, Cary, NC) and Stata version 7.0 (Stata Corp, College Station, TX).
Results The final study cohort included 1587 patients with Crohn’s disease (62% female; mean age at start of follow-up, 41 ⫾ 17 years) and 2773 patients with ulcerative colitis (50% female; mean age at start of followup, 48 ⫾ 16 years) (Table 1). The median duration from the last prescription for inflammatory bowel disease therapies to the start of a flare was 247 days (interquartile range, 175 to 442 days) for Crohn’s disease and 243 days (interquartile range, 172 to 440 days) for ulcerative colitis. Identification of Flare of Inflammatory Bowel Disease Completed surveys with usable data were received from GPs on 136 of the 150 patients (91%). The 136 patients (79 female and 57 male) included 59 with Crohn’s disease, 68 with ulcerative colitis, 5 with inflammatory bowel disease not otherwise specified, and 4 in whom the GP reported that the patient did not have inflammatory bowel disease. The mean age on the index
668
LEWIS ET AL.
GASTROENTEROLOGY Vol. 126, No. 3
Table 2. Odds Ratio for the Association Between Season of the Year and Flare of Disease Crohn’s disease
Winter Spring Summer Autumn
Ulcerative colitis
Model 1 OR (95% CI)
Model 2 OR (95% CI)
Model 1 OR (95% CI)
Model 2 OR (95% CI)
Reference 1.06 (0.95–1.18) 1.07 (0.96–1.19) 1.05 (0.94–1.18)
Reference 1.02 (0.86–1.21) 1.04 (0.87–1.23) 1.07 (0.90–1.27)
Reference 1.13 (1.05–1.23) 1.07 (0.99–1.16) 1.08 (0.99–1.17)
Reference 1.14 (1.01–1.28) 1.07 (0.94–1.21) 1.06 (0.94–1.20)
NOTE. Model 1: Logistic regression model adjusted for calendar year and sex. Model 2: Conditional logistic regression model adjusted for calendar year, aspirin, NSAID, azathioprine/6-mercaptopurine, and antibiotic use.
date was 47 ⫾ 20 years. After excluding patients with a previous diagnosis of rheumatoid arthritis, chronic obstructive airway disease, or asthma, the positive predictive value of a new prescription for steroids to identify an acute flare of inflammatory bowel disease was 82 of 96 (85%). We used each patient’s multiple episodes of drug use to examine whether a 4-month window was sufficient to identify a patient no longer receiving steroids. Using this algorithm, the positive predictive value of no longer taking steroids 4 months after the last prescription was 29 of 32 (91%). To evaluate whether prescriptions identifying a flare might be routine refills for patients in remission, we examined the timing of prescriptions before and after a flare. The first interval was the median time from the last prescription for steroids or 5-ASA therapy until a new flare was identified. The second interval was the median time from the prescription identifying a flare to the next prescription for steroids or 5-ASA. The interval between prescriptions was markedly shorter after the flare for both Crohn’s disease (median, 183 days [interquartile range, 143 to 312 days] before a flare vs. 71 days [interquartile range, 33 to 127 days] after a flare, P ⬍ 0.0001) and ulcerative colitis (median, 180 days [interquartile range,
Figure 1. Seasonality of flares within the cohort of Crohn’s disease patients.
141 to 297 days] before a flare vs. 84 days [interquartile range, 36 to 140 days] after a flare, P ⬍ 0.0001). This suggests that prescriptions identifying a flare are unlikely to be routine refills for patients in remission. Association of Season of the Year and Flare Within the Cohort of Inflammatory Bowel Disease Patients In the logistic regression model adjusting only for calendar year and sex, there was no association between season of the year and flare of Crohn’s disease (P ⫽ 0.66 for overall comparison of all seasons) (Table 2 and Figure 1). For ulcerative colitis, the model showed a very weak but statistically significant association between season of the year and flare of disease (P ⫽ 0.02) (Figure 2). Using winter as the reference category, only spring had a higher rate of flare of disease (OR ⫽ 1.13, 95% CI: 1.05–1.23). The conditional logistic regression models adjusting for calendar year, NSAID, aspirin, azathioprine, and antibiotic exposure produced nearly identical results (Table 2). Sensitivity analyses were performed in which flares were identified only by a prescription for a 5-ASA product or only a prescription for steroids using the fully
Figure 2. Seasonality of flares within the cohort of ulcerative colitis patients. Using winter as the reference category, only spring had a slightly higher rate of flare of disease (OR ⫽ 1.13, 95% CI: 1.05– 1.23).
March 2004
SEASONALITY OF IBD
Table 3. Sensitivity Analyses Examining Different Definition of Flare of Crohn’s Disease 5-ASA or steroid OR (95% CI) Winter Spring Summer Autumn
5-ASA only OR (95% CI)
Table 4. Sensitivity Analyses Examining Different Definition of Flare of Ulcerative Colitis
Steroid only OR (95% CI)
Reference Reference Reference 1.02 (0.86–1.21) 0.99 (0.79–1.23) 1.06 (0.83–1.35) 1.04 (0.87–1.23) 1.09 (0.88–1.36) 0.96 (0.75–1.21) 1.07 (0.90–1.27) 1.03 (0.83–1.29) 1.20 (0.94–1.53)
adjusted conditional logistic regression models. In all cases, although the confidence intervals are wider because of reduced number of observations, similar results were obtained in each model (Tables 3 and 4). Likewise, sensitivity analyses were conducted in which we excluded patients with multiple flares, at least one of which occurred on the same date as a diagnosis for a potentially noninflammatory bowel disease indication for steroids (n ⫽ 577 for Crohn’s disease and n ⫽ 989 for ulcerative colitis). These analyses again resulted in nearly identical results (data not shown). A second set of sensitivity analyses was conducted in which we required longer periods without a prescription for steroids or 5-ASA products to identify new flares of disease in our model adjusted for calendar year and sex. Increasing this time period from 4 months to 6 months or 8 months had a minimal effect on the results. The greatest absolute change in the estimated odds ratio was 0.11 (risk of flare of ulcerative colitis in the spring increased from 1.13 to 1.24 in the 8-month model). In no case did any nonstatistically significant results become statistically significant. In another sensitivity analysis, we required a minimum of 2 prescriptions for steroids or 5-ASA products within a 60-day window to identify a new flare. Again, the results were nearly identical to the primary analyses. The largest absolute difference in the odds ratio from the primary analyses was 0.08 (risk of flare of Crohn’s disease in the spring increased from 1.06 to 1.14). Additional sensitivity analyses were performed using a Monte Carlo simulation to examine the potential effect of misclassification of the dates of flares. The mean odds ratios based on 500 simulations, in which random adjustment forward was made to the date of flare according to the half-normal distribution, were very close to our original estimates (Table 5). In addition, the distribution of the odds ratios from the Monte Carlo simulation suggested that misclassification of the date of a flare was unlikely to affect substantially our conclusions. The largest absolute difference between the odds ratio from our primary analysis and the upper bound of the range of odds ratios from the simulation was only 0.12 (Crohn’s disease flare in spring).
669
5-ASA or steroid OR (95% CI) Winter Spring Summer Autumn
5-ASA only OR (95% CI)
Steroid only OR (95% CI)
Reference Reference Reference 1.14 (1.01–1.28) 1.14 (0.99–1.30) 1.09 (0.88–1.34) 1.07 (0.94–1.21) 1.05 (0.92–1.20) 0.99 (0.80–1.22) 1.06 (0.94–1.20) 1.04 (0.91–1.19) 1.10 (0.89–1.36)
Association of Season of the Year and Flare Within Individual Inflammatory Bowel Disease Patients To determine whether individual patients tend to have flares during the same season of the year, we examined subgroups of the cohorts who experienced more than 1 flare. Among those patients with 2 flares, 3 flares, and 4 flares, the concurrence of multiple flares within a single season was no greater than expected for either Crohn’s disease or ulcerative colitis (P ⬎ 0.05 for all comparisons) (Table 6 and Figures 3 and 4). Of note, although this analysis approached statistical significance in patients with 2 flares, this was due to these flares occurring less commonly than expected in the same season. To further assess the relationship between season of flares within individual subjects, we fit a version of our GEE logistic regression model using autoregressive correlational structure (AR4). The model parameters and associated P values were not significantly impacted by the change in correlational structure applied by using the autoregressive structure. The correlations of importance that are estimated in the model are the correlations between flare status for 2 seasons separated by 1 season, 2 seasons, 3 seasons, and 1 year. If individual patients tended to flare during the same season, we would expect to see higher correlation coefficients in the fourth season after the initial flare. In fact, the correlation coefficients were comparable among the second, third, and fourth Table 5. Comparison of Results From the Primary Analyses to the Monte Carlo Simulation
Season Crohn’s disease Spring Summer Autumn Ulcerative colitis Spring Summer Autumn
Odds ratio from primary analyses
Mean odds ratio from Monte Carlo simulation
Range of odds ratios from Monte Carlo simulation
1.06 1.07 1.05
1.09 1.08 1.04
1.00–1.18 0.99–1.16 0.95–1.14
1.13 1.07 1.08
1.08 1.05 1.02
1.02–1.15 1.01–1.10 0.96–1.08
670
LEWIS ET AL.
GASTROENTEROLOGY Vol. 126, No. 3
Table 6. Periodicity of Flares Within Individual Subjects Number of flares per subject
Crohn’s disease patients (n)
Crohn’s disease P value
Ulcerative colitis patients (n)
Ulcerative colitis P value
2 3 4
266 120 74
0.06 0.23 0.25
506 242 130
0.09 0.64 0.48
seasons following the flare (0.05 to 0.06 for ulcerative colitis and 0.03 for all 3 seasons for Crohn’s disease). These results provide further support that individual patients do not tend to flare during the same season.
Discussion There are several reasons to hypothesize that inflammatory bowel disease may have a seasonal pattern. Environmental exposures vary with seasons of the year. These environmental exposures could influence the natural history of inflammatory bowel disease. For example, use of medications that may influence the natural history of inflammatory bowel disease, such as NSAIDs,18 –20 may vary across seasons. Perhaps more importantly, exposure to antigens such as viral infections vary according to season of the year.21,22 Nearly 70 years ago, Banks and Bargen suggested that upper respiratory tract infection may predispose patients to exacerbation of otherwise quiescent ulcerative colitis.23 They noted these infections in 60% of patients who relapsed. Similar results have been reported by a few other investigators,9,24 whereas others have failed to identify an association between upper respiratory tract infection and relapse of inflam-
Figure 3. Time from first flare to second flare among patients with Crohn’s disease who experienced more than 1 flare. Each bar represents a period of 100 days. The first bar represents the period from day 101 to 200. There is no bar for the period from day 1 to 100 because our algorithm required a minimum of 122 days between prescriptions to identify a new flare.
Figure 4. Time from first flare to second flare among patients with ulcerative colitis who experienced more than 1 flare. Each bar represents a period of 100 days. The first bar represents the period from day 101 to 200. There is no bar for the period from day 1 to 100 because our algorithm required a minimum of 122 days between prescriptions to identify a new flare.
matory bowel disease.4 These studies have generally been limited by small sample sizes and in some cases by inability to determine exact temporal relationships between upper respiratory tract infection and relapse of inflammatory bowel disease. Nonetheless, they provide reason to believe that relapse of inflammatory bowel disease could be seasonal. Previous research has resulted in conflicting opinions as to whether inflammatory bowel disease is more likely to flare during selected periods of the year (Table 7).1–9 Two prior studies demonstrated increased rates of ulcerative colitis flares in the spring,6,9 whereas other studies have suggested increased rates of flares in the autumn or winter.1,2,4 Still other studies have suggested that there is no pattern of flare among patients with ulcerative colitis.3,7 The results for Crohn’s disease have been equally as inconsistent.1,3,5,7,9 In addition, patients frequently report that their disease tends to be more active during certain periods of the year. Our study does not support an association between season of the year and flares of disease either within the cohort or within individual patients with inflammatory bowel disease. Although there was a statistically significant increase in the risk of flare of ulcerative colitis in the spring, the magnitude of this association was extremely small (OR ⫽ 1.14). Given the numerous statistical tests performed, it is likely that this was a chance finding (i.e., type 1 error). Furthermore, if there is an association between season of the year and flare of disease, we would have expected to also see evidence of this association among individual patients with multiple
March 2004
SEASONALITY OF IBD
671
Table 7. Seasonal Pattern of IBD Exacerbation Author and year 1
Myszor and Calam, 1984 Sellu, 19862 Don and Goldacre, 19843 Riley et al., 19904 Zeng and Anderson, 19965 Tysk and Jarnerof, 19936 Sonnenberg et al., 19947 Sonnenberg and Wassermann, 19918 Kangro et al., 19909a aCD
Location
CD peak
UC peak
United Kingdom United Kingdom United Kingdom United Kingdom Canada Sweden USA USA
No pattern Not reported No pattern Not reported September–February Not Reported No pattern Pattern similar to all other diagnoses
September–February September–February No pattern August–January Not reported Spring and Autumn No pattern Pattern similar to all other diagnoses
United Kingdom
Winter, April–May, and July–August
Winter, April–May, and July–August
and UC data were combined in this study.
flares. In contrast, when we limited our analyses to patients with multiple flares, we saw no evidence of a seasonal pattern of either Crohn’s disease or ulcerative colitis. Thus, we feel that these data argue against a seasonal pattern to the natural history of inflammatory bowel disease. Our study has several features that make it unique compared with the prior studies. Most notable is that a large cohort of patients was followed for an extended period of time, such that we could account for the effect of multiple relapses among an individual patient. Also, to our knowledge, this is the first study to examine the periodicity of relapse within individual subjects. Another unique feature of our study was the ability to adjust for changes in medication exposures over the course of multiple flares and remissions. Medications such as antibiotics and NSAIDs could potentially influence relapse rates of inflammatory bowel disease. Failure to adjust for these medications could bias results of studies addressing seasonality of relapses. Because the GPRD contains data on medications prescribed, we were able to account for the variable use of medications across the course of the study. Our data were from the records of primary care physicians, not gastroenterologists. As such, this population likely contains more patients with mild disease than is seen in referral centers. Importantly, we have previously validated the diagnosis of inflammatory bowel disease within the GPRD and demonstrated that the prevalence, anatomic distribution of disease, and prescribing patterns for inflammatory bowel disease patients within GPRD are consistent with that observed in other populations.16 Furthermore, in this study, we performed additional validation studies to examine the reliability of using medication prescriptions to identify exacerbations of inflammatory bowel disease. Our results demonstrated relatively high predictive values for steroid prescriptions to identify flares of inflammatory bowel disease. Importantly, however, we used new 5-ASA prescriptions to
identify flares as well. We elected not to validate the new 5-ASA prescriptions as a marker of disease relapse, in part because of financial constraints. Given that these medications are much more specific for inflammatory bowel disease than are steroids, the predictive value of a new 5-ASA prescription to identify a disease relapse should be even higher than that of a new steroid prescription. Any nondifferential misclassification of remission periods as flares would be expected to bias our results toward the null. However, and importantly, when we reanalyzed our data using only steroids or only 5-ASA medications to identify flares of disease, we obtained nearly identical results. As such, we feel confident that our algorithm is not significantly biased by including both 5-ASA and steroids as a marker of relapse. It is possible that a small proportion of flares may have gone undocumented, either because the symptoms were so mild that the patients did not seek treatment or if the flare was treated by other physicians. Within our validation study, we documented that more than 83% of the patients with Crohn’s disease and 80% of those with ulcerative colitis were receiving care from a gastroenterologist as well as from their GP (data not shown). However, several features of our study make this unlikely to have significantly biased our results. First, if flares went completely unrecorded because they were treated exclusively by the gastroenterologist, this pattern would need to occur preferentially in the seasons with the highest rates of flare to mask a true association. There is no reason to believe that flares in one season of the year would be more likely to be treated exclusively by a gastroenterologist than flares in other seasons of the year. Second, within the United Kingdom, specialists usually prescribe a small quantity of medication, and refills are written by the GP. All of our patients had received prescriptions for inflammatory bowel disease medications from their GP as a requirement for inclusion. Thus, these GPs are accustomed to writing for steroids and 5-ASA products. In addition, if some flares were initially treated
672
LEWIS ET AL.
by the gastroenterologist, it is expected that the GP would write the continuing prescriptions for these medications. Although this may slightly change the estimated date of the flare, our Monte Carlo simulation demonstrates that it would be unlikely to substantially change the results. Thus, we believe that misclassification related to identification and dating the start of flares is unlikely to have significantly affected our results. It is possible that a few of the prescriptions that we used to identify a flare could actually have been a refill for a patient in remission. Importantly, when we used even more stringent definitions of a flare (e.g., an 8-month interval without prescriptions for steroids or 5-ASA products), we obtained nearly identical results. Thus, this form of misclassification is unlikely to have significantly influenced our results. Another potential limitation of our data is the possibility of incomplete adjusting for confounders. Although we were able to adjust for many factors potentially associated with relapse of disease, there may be other factors not accounted for in our analyses. Our case crossover analyses matching each patient to herself or himself as a control should minimize the effect of potential confounders that are unlikely to vary across time, such as smoking habits. We elected to exclude patients with rheumatoid arthritis, asthma, and chronic obstructive airway disease from our analyses so as to improve the specificity of our algorithm to identify flares of inflammatory bowel disease. It is possible that the subgroup of patients with both inflammatory bowel disease and reactive airway disease are uniquely sensitive to seasonal variations. However, when we include those subjects in the analyses, we obtain very similar results for both within cohort and within subject comparisons (data not shown). Likewise, we excluded patients with less than 1 year of follow-up and started our follow-up time at least 1 year after the first diagnosis with inflammatory bowel disease. This was done to avoid potential bias associated with newly diagnosed patients or from inclusion of patients who received only short-term care from the general practitioner. Importantly, when we repeated the analyses including these patients, the results were again nearly identical (data not shown). As such, it seems unlikely that exclusion of either of these groups of patients has biased our results. We excluded patients who were chronically maintained on steroids and/or 5-ASA medications. This was necessary to be able to identify a flare of disease. It is possible, albeit unlikely, that such patients would have a different seasonal pattern of relapse than patients who
GASTROENTEROLOGY Vol. 126, No. 3
discontinue these medications. Unfortunately, we are unable to address this question within this cohort. In conclusion, our data do not support a seasonal pattern to the natural history of inflammatory bowel disease. We found little evidence of seasonality of relapse within a cohort of patients with inflammatory bowel disease or within individual inflammatory bowel disease patients. Although our data do not exclude the possibility that environmental exposures such as upper respiratory tract infections or their treatments may trigger a relapse of inflammatory bowel disease, it seems less likely that environmental exposures that vary with season of the year explain a large proportion of the variance in the natural history of this disease.
References 1. Myszor M, Calam J. Seasonality of ulcerative colitis. Lancet 1984;2:522–523. 2. Sellu DP. Seasonal variation in onset of exacerbations of ulcerative proctocolitis. J Royal Col Surg Edinburgh 1986;31:158 – 160. 3. Don BA, Goldacre MJ. Absence of seasonality in emergency hospital admissions for inflammatory bowel disease. Lancet 1984;2:1156 –1157. 4. Riley SA, Mani V, Goodman MJ, Lucas S. Why do patients with ulcerative colitis relapse? Gut 1990;31:179 –183. 5. Zeng L, Anderson FH. Seasonal change in the exacerbations of Crohn’s disease. Scand J Gastroenterol 1996;31:79 – 82. 6. Tysk C, Jarnerot G. Seasonal variation in exacerbations of ulcerative colitis. Scand J Gastroenterol 1993;28:95–96. 7. Sonnenberg A, Jacobsen SJ, Wasserman IH. Periodicity of hospital admissions for inflammatory bowel disease. Am J Gastroenterol 1994;89:847– 851. 8. Sonnenberg A, Wasserman IH. Epidemiology of inflammatory bowel disease among U.S. military veterans. Gastroenterology 1991;101:122–130. 9. Kangro HO, Chong SK, Hardiman A, Heath RB, Walker-Smith JA. A prospective study of viral and mycoplasma infections in chronic inflammatory bowel disease. Gastroenterology 1990;98:549 – 553. 10. Lis Y, Mann RD. The VAMP Research multi-purpose database in the U.K. J Clin Epidemiol 1995;48:431– 443. 11. Anonymous. The General Practice Research Database: Information for Researchers. Office for National Statistics, 1994. 12. Garcia Rodriguez LA, Perez-Gutthann S. Use of the UK General Practice Research Database for pharmacoepidemiology. Br J Clin Pharmacol 1998;45:419 – 425. 13. Jick H, Jick SS, Derby LE. Validation of information recorded on general practitioner based computerised data resource in the United Kingdom. BMJ 1991;302:766 –768. 14. Jick H, Terris B, Derby L, Jick S. Further validation of information recorded on a general practitioner based computerized data resource in the United Kingdom. Pharmacoepidemiol Drug Safety 1992;1:347–349. 15. Garcia Rodriguez L, Perez-Gutthann S, Jick S. The UK General Practice Research Database. In: Strom B, ed. Pharmacoepidemiology. 3rd ed. New York: John Wiley & Sons, 2000:375–385. 16. Lewis JD, Brensinger C, Bilker WB, Strom BL. Validity and completeness of the General Practice Research Database for studies of inflammatory bowel disease. Pharmacoepidemiol Drug Safety 2002;11:211–218. 17. Maclure M. The case-crossover design: a method for studying
March 2004
18.
19.
20.
21.
transient effects on the risk of acute events. Am J Epidemiol 1991;133:144 –153. Gibson GR, Whitacre EB, Ricotti CA. Colitis induced by nonsteroidal anti-inflammatory drugs. Report of four cases and review of the literature. Arch Intern Med 1992;152:625– 632. Evans JM, McMahon AD, Murray FE, McDevitt DG, MacDonald TM. Non-steroidal anti-inflammatory drugs are associated with emergency admission to hospital for colitis due to inflammatory bowel disease. Gut 1997;40:619 – 622. Felder JB, Korelitz BI, Rajapakse R, Schwarz S, Horatagis AP, Gleim G. Effects of nonsteroidal antiinflammatory drugs on inflammatory bowel disease: a case-control study. Am J Gastroenterol 2000;95:1949 –1954. Dowell SF. Seasonal variation in host susceptibility and cycles of certain infectious diseases. Emerg Infect Dis 2001;7:369 –374.
SEASONALITY OF IBD
673
22. Noah ND. Cyclical patterns and predictability in infection. Epidemiol Infect 1989;102:175–190. 23. Banks B, Bargen J. Relapses in chronic ulcerative colitis— causes and prevention. Arch Intern Med 1934;53:131–139. 24. Mee AS, Jewell DP. Factors inducing relapse in inflammatory bowel disease. BMJ 1978;2:801– 802.
Received June 2, 2003. Accepted December 4, 2003. Address requests for reprints to: James D. Lewis, M.D., MSCE, University of Pennsylvania, Center for Clinical Epidemiology and Biostatistics, 7th Floor Blockley Hall, 423 Guardian Drive, Philadelphia, Pennsylvania 19104-6021. e-mail: [email protected]; fax: (215) 573-5325. Supported by National Institutes of Health grant K08-DK02589.