Baseline Predictors of Discontinuation of Prescription Drug Therapy for IBS-C: Cohort Analysis at an Integrated Healthcare System.

We analyzed a cohort of outpatients seen across the Michigan Medicine integrated health system, which includes a large regional network of primary and specialty care offices across southeast Michigan. Our study followed the ISPOR checklist for medication persistence studies [ 16 ] and STROBE checklist for cohort studies. The Institutional Review Boards at Michigan Medicine and Dartmouth-Hitchcock determined that this study was exempt from review, prior to initiation of our study. Individuals were identified using the Michigan Medicine Electronic Medical Record Search Engine (EMERSE) which links encounter information, administrative claims, and prescription data across the integrated healthcare system [ 17 ]. We included adults at least 18 years of age with IBS-C (defined using a validated set of International Classification of Diseases [ICD]-9 codes 564.1 or 388.*[ 18 ]) who started linaclotide or lubiprostone between December 2012 (i.e., when linaclotide received approval notification by the U.S. Food and Drug Administration [FDA] [ 19 ]) and November 2016. These individuals were followed until treatment discontinuation and excluded if they did not have at least one office visit subsequent to initiating therapy. We did not perform assessments of plecanatide, tegaserod, or tenapanor as these agents were neither FDA approved nor commercially available at the time of this study. Age at treatment initiation and sex were extracted as patient factors, and specialty of the managing provider was extracted as a provider factor (classified as gastroenterology/non-gastroenterology). The presence of individual COPCs and mood disorders was defined using sets of ICD-9/10 codes contained in the patient history or in outpatient problem lists/diagnoses within the preceding 12 months before initiation of IBS-C treatment ( Supplement Table 1 ). Use of several medication classes common among patients with chronic pain were extracted within the 90-day preceding IBS-C treatment initiation, including non-opioid pain medications (acetaminophen or nonsteroidal anti-inflammatory drugs [NSAID]), narcotics, and benzodiazepines. The primary outcome of our study was medication persistence, which is defined as “the act of continuing the treatment for the prescribed duration” [ 20 ]. Charts were reviewed manually in the original study detailing the medication persistence outcome measure, including all office notes, phone notes, and outside records. We recently identified reasons for IBS-C treatment discontinuation (i.e., loss of medication persistence in IBS-C populations) in a separate study; the most common reasons were loss of treatment efficacy, insurance non-coverage, and adverse events related to treatment [ 7 ]. Medication persistence was previously and rigorously extracted in manual chart review and defined by the number of days between date of treatment initiation and discontinuation, consistent with the definition of medication persistence provided by the ISPOR Medication Adherence and Persistence Special Interest Group [ 20 ]. Indicator variables were created to classify individual COPCs, mood disorders, concurrent medications, presence of IBS diagnosis, and sex. Continuous variables were created for age as well as for number of days covered as our measure of therapeutic persistence. Means and standard deviation (SD) were used to represent categorical variables, and the Fischer χ 2 and Student’s t-test were used to represent continuous variables. A Cox proportional hazards model was constructed to predict medication persistence based on age, sex, individual COPCs, at least one COPC, individual mood disorders or at least one mood disorder, and provider specialty. Censoring was defined as remaining on prescription beyond the most recent patient encounter. To account for these missing data, our analysis only included the time period during which a patient was known to have remained on therapy. Analyses were stratified by IBS-C drug. Univariate analyses were performed for each covariate in the model. A forward stepwise selection method was used to construct the final multivariable model, selecting variables with at least a p -value <0.10 in univariate analyses and advancing age and sex into the final model. This model was validated using a backward stepwise Cox model construction. Goodness of fit was assessed by plotting Cox–Snell residuals. Statistical analysis was conducted using STATA version 15.0 (STATACorp, College Station, TX).

Our overall cohort included 225 IBS-C patients taking linaclotide and 492 taking lubiprostone who met eligibility criteria. The mean age of individuals was 48.3 years, 86.9% were women, 46.6% had at least one COPC, and 60.3% had at least one diagnosed mood disorder. The average length of follow-up in our cohort was 2.1 years (SD 1.6 years). In the linaclotide cohort, 45 patients (20.0%) discontinued treatment within 3 months and 68 patients (32.2%) discontinued within one year. In the lubiprostone cohort, 120 patients (24.4%) discontinued treatment within 3 months and almost half of patients (225 patients, 47.7%) discontinued within one year. Use of narcotic pain medications, non-narcotic pain medications, and benzodiazepines was common in our overall cohort. Almost one-half of patients had used at least once dose of a narcotic pain medication, and almost one-third had used at least one dose of a benzodiazepine within the 90 days prior to initiating IBS-C treatment. Half of the patients were prescribed acetaminophen or NSAIDs for management of chronic pain. There were no statistically significant differences between the linaclotide and lubiprostone cohorts with respects to age, sex, baseline frequency of COPCs and mood disorders, or use of non-narcotic pain medications. Baseline use of narcotics in the preceding 90 days was higher in the lubiprostone cohort compared to the linaclotide cohort, consistent with FDA labeling for lubiprostone in treating opioid-induced constipation for chronic non-cancer pain ( Table 1 ). Women were less likely to discontinue linaclotide compared to men ( p = 0.028), while sex did not affect treatment outcomes for lubiprostone ( p = 0.93). Age of the patient at treatment onset did not influence the risk of discontinuation, nor did specialty of the prescribing/managing provider. The predictive values of baseline comorbidities on treatment discontinuation are reported in Table 2 for linaclotide and Table 3 for lubiprostone. The presence of at least one COPC reduced the risk of discontinuing linaclotide (hazard ratio [HR] = 0.57, 95% confidence interval [CI] 0.37–0.86, p = 0.008) but did not influence the risk of discontinuing lubiprostone (HR = 1.25 [0.997–1.58], p = 0.053). Among individual COPCs, osteoarthritis (HR = 1.66 [1.24–2.21], p = 0.001) and chronic pelvic pain/vulvodynia (HR = 1.96 [1.27–3.05], p = 0.003) increased the risk of discontinuing lubiprostone. Comorbid depression and anxiety did not significantly influence the risk of discontinuing either treatment. Findings were stratified solely on the presence of at least one COPC; in other words, having more than one COPC did not further modify treatment outcomes. Patients with at least two COPCs had a similar risk of drug discontinuation over time in the linaclotide cohort (HR = 1.12 [0.60–2.10], p = 0.725) and the lubiprostone cohort (HR = 0.99 [0.71–1.39], p = 0.967) compared to patients with only one diagnosed COPC. Patients who were prescribed non-narcotic pain medications (acetaminophen or NSAIDs) during the preceding 90 days had a greater risk of discontinuing lubiprostone (HR = 1.34 [1.07–1.69], p = 0.012) compared to patients who did not report use of these medications. This effect was not seen in the linaclotide cohort. Use of narcotic pain medications or benzodiazepines did not significantly influence the risk of discontinuing lubiprostone or linaclotide. Ninety-five patients in the lubiprostone cohort (19.3% of the total cohort) were treatment-experienced to linaclotide, while 152 patients on linaclotide had previously tried lubiprostone (67.6% of the total cohort). In the linaclotide cohort, prior exposure to IBS-C prescription drug therapies did not significantly affect discontinuation rates on linaclotide (HR = 0.78 [0.51–1.20]). In contrast, linaclotide-experienced patients had a higher risk of discontinuing lubiprostone compared to treatment-naïve patients (HR = 1.64, [1.25–2.15]). We first categorized COPCs broadly by defining patients as having at least one COPC, or none at all. Whereas the presence of at least one COPC decreased the risk of discontinuing linaclotide (HR = 0.58 [0.39 = 0.89], p = 0.013), it had no significant impact on the risk of discontinuing lubiprostone. However, non-narcotic pain medications at baseline (HR = 1.33 [1.05–1.69], p = 0.018) increased the risk of discontinuing lubiprostone, an effect not seen with linaclotide. Age- and sex-adjusted survival curves representing medication persistence on linaclotide and lubiprostone are presented in Fig. 1 , stratified on the presence of comorbid COPCs. We then assessed COPCs individually. The presence of migraine headaches (HR = 0.56 [0.32–0.99], p = 0.047) or chronic pelvic pain/vulvodynia (HR = 0.35 [0.12–0.98], p = 0.047) decreased the risk of discontinuing linaclotide, while the presence of chronic fatigue syndrome (HR = 1.99 [1.07–3.69], p = 0.030) increased the risk of discontinuing linaclotide. In contrast, chronic pelvic pain/vulvodynia increased the risk of discontinuing lubiprostone (HR = 1.78 [95% CI 1.14–2.78], p = 0.011). The final age- and sex-adjusted models are presented in Tables 2 and 3 ; findings were consistent with univariate analyses. Comorbid mood disorders, use of narcotic pain medications, and use of benzodiazepines did not significantly influence the risk of treatment discontinuation and were not included in the final predictive models.

Within a large integrated healthcare system, we identified key determinants of drug discontinuation associated with the use of two commonly prescribed secretagogues for the management of IBS-C. The presence of comorbid COPCs (specifically chronic pelvic pain/vulvodynia and migraine headaches) reduced the risk of discontinuing linaclotide, while comorbid chronic pelvic pain/vulvodynia increased the risk of discontinuing lubiprostone. Patients prescribed non-narcotic pain medications for chronic pain management were also at greater risk of discontinuing lubiprostone. Importantly, mood disorders or the use of narcotic pain medications or benzodiazepines did not appear to influence the risk of IBS-C drug discontinuation. Prior IBS-C treatment exposure appears important for patients considering lubiprostone, but treatment experience did not significantly increase the risk of discontinuing linaclotide. These determinants (and non-determinants) are common comorbidities in IBS populations and are readily identifiable in electronic health records. Due to significant efforts over the past two decades, there are now eight FDA-approved drugs for IBS. To help providers identify appropriate therapy, IBS is subtyped by predominant stool texture: constipation, diarrhea, or mixed. Yet within this framework, there still remain three FDA-approved IBS-D (rifaximin, eluxadoline, alosetron) and five FDA-approved IBS-C treatments (linaclotide, plecanatide, lubiprostone, tegaserod, tenapanor). Multiple treatment options can result in choice overload for patients and healthcare providers (assuming comparable availability and insurance coverage) especially when a lack of head-to-head trials limits clinical comparisons [ 21 - 23 ]. Importantly, these treatments represent more than just anti-diarrheals and laxatives; IBS treatments target underlying disease mechanisms including small intestinal bacterial overgrowth and visceral hypersensitivity [ 3 ]. Efforts to phenotype patients based on expected disease responses, as is done in inflammatory bowel disease, may facilitate more effective treat-to-target approaches as compared to current empirical IBS management strategies [ 8 , 24 ]. Our findings are consistent with visceral hypersensitivity as an important pathophysiologic mechanism in IBS-C and may suggest further study to rigorously evaluate the biological plausibility of discrepant effects of COPCs on medication persistence. That comorbid COPCs appear to reduce the risk of discontinuation on linaclotide supports a mechanism of action which links guanylate cyclase-C agonism with reduced colonic nociception [ 25 ]; whether linaclotide (or lubiprostone) is effective in managing non-GI COPCs (vs. an IBS patient with evidence of a visceral hypersensitivity pathophysiology evidenced by having several overlapping pain conditions) remains unknown. Further efforts to identify differences in serum-based biological markers such as bile-acid precursors, functional MRI studies, volatile organic compounds, hydrogen/methane breath testing, and small bowel aspirates may ultimately augment the ability of providers to tailor therapy for this complex and heterogenous disease [ 24 ]. Narcotic, benzodiazepine use, and mood disorders are commonly identified in IBS populations—these findings were reaffirmed in over half of patients in our cohort. Providers often perceive that outcomes for IBS patients with these comorbidities will be poorer [ 26 ]; however, this perception is not supported by our findings. Patients taking narcotics or benzodiazepines, and patients with comorbid mood disorders, did not experience higher risks of treatment discontinuation compared to IBS-C patients lacking these factors. Reassuringly, these patients appear to do just as well as patients without these conditions with regard to the length of time that patients remain on safe and effective IBS-C drug treatments. Interestingly, patients who are managed with prescription non-narcotic pain medications by a provider (defined as acetaminophen and NSAIDs) appear to be at higher risk of discontinuing lubiprostone even after controlling for comorbid COPCs. It is important to recognize that acetaminophen and NSAIDs are particularly effective for somatic pain (as opposed to visceral pain in IBS) [ 27 ]. In fact, abdominal wall pain is a common referral to gastroenterology which causes somatic-type pain and can be diagnosed by evaluating for a Carnett sign on physical examination [ 28 ]. In practice, clinicians should pay careful attention to the broad differential for abdominal pain which can be gleaned by a careful history and physical examination to differentiate IBS from other causes of abdominal pain. A detailed and practical review of these causes is available in a recent edition of the American Journal of Gastroenterology by Pichetshote et al. [ 29 ]. One major limitation of our study is that it was performed at a single integrated healthcare system and that we only assessed the two IBS-C treatments available at the time, recognizing that more recently approved agents (plecanatide, tegaserod, tenapanor) may not yet have sufficient longitudinal data to facilitate meaningful analysis of real-world medication persistence. Another limitation relates to significant differences in baseline patient characteristics between linaclotide and lubiprostone cohorts, recognizing the observational nature of our study as compared to a head-to-head trial. Patients starting linaclotide were more likely to take concurrent narcotics or TCA/SSRI and also less likely to be treatment-naïve to IBS-C drugs as compared to patients started on lubiprostone. We addressed these differences by including these variables in multivariable regression analysis, noting that definitive conclusions on comparative efficacy are outside the scope of this study and should be limited in this exploratory analysis focused on the intersection between COPC and IBS paradigms. As with any longitudinal large-scale claims-based analysis, several other important limitations should be considered. IBS is a symptom-driven diagnosis with few biological markers in clinical practice, and the IBS-specific patient-reported outcome measures which are routinely used in clinical trials are not commonly assessed in general populations [ 30 ]. Thus, efforts to use large-scale datasets to phenotype IBS treatment response remain challenging outside of drug trials (in which evaluations of complex phenotypes using either standard regression methods or machine learning remains limited by study enrollment and power) [ 31 ]. To overcome these barriers, we used medication persistence (i.e., time until treatment discontinuation) as an important, valid, and quantifiable surrogate biomarker, recognizing that treatment discontinuation in IBS only rarely represents improvement of the underlying illness [ 7 ]. Furthermore, we defined IBS and comorbidities using administrative claims data; all of these comorbidities are defined in clinical practice using patient-reported outcomes. At the same time, the use of PROs to evaluate for each specific COPC is not feasible in practice due to the length of PROs and ownership of each COPC by a different specialty. The claims data we used are similar to what is available to general gastroenterologists in usual practice. Understanding the limitations of using claims data due to mis-coding, it is important to recognize the high threshold of labeling patients with a mood disorder or COPC (such as chronic pelvic pain/vulvodynia, fibromyalgia, or endometriosis) in standard practice and that these conditions are likely under-recognized (rather than over-recognized) [ 26 ]. It is also possible that our results may be subject to confounding due to over-the-counter supplements (including a number of antispasmodics and laxatives). While differences between drug cohorts were found with regard to use of several medication classes ( Table 1 ), analyses on medication classes did not reveal significant differences in medication persistence between IBS-C drugs (data not shown). We performed analyses mainly on diagnosis coding rather than on concurrent medications, because adherence to concurrent medications could be adequately measured. Recognizing the observational nature of our study, our population was not systematically characterized for the presence of functional defecatory disorders which are attributed as an etiology of symptoms up to 40% of individuals with laxative-refractory chronic constipation [ 32 ]. While it is possible that poorly managed mood disorders might underlie COPCs, we did not detect interactions between mood disorders and COPCs. Overall, our findings are exploratory in nature, and further prospective studies are required to confirm the observations seen in our study; pathophysiologic relationships among COPCs (including IBS) have the potential to yield biomarkers on which to stratify treatment response. A design of such study could follow the basic outline of the CONTOR study following patients with IBS-C or chronic idiopathic constipation containing a rigorous set of patient-reported outcomes alongside linked pharmacy and medical claims data [ 33 ]. In prescribing linaclotide and lubiprostone in practice, comorbid COPCs may influence the risk of discontinuing secretagogues commonly used to treat IBS-C. Patients with chronic pelvic pain/vulvodynia appear to be at higher risk of discontinuing lubiprostone. Patients with chronic pelvic pain/vulvodynia, migraine headache, or at least one COPC appear to be at lower risk of discontinuing linaclotide which may relate to direct effects of GC-C agonism on abdominal pain sensation related to a global visceral hypersensitivity phenotype of IBS. Use of non-narcotic pain medications for somatic pain (acetaminophen and NSAIDs) under the care of a managing provider increases the risk of discontinuing lubiprostone. Patients with comorbid mood disorders or those using narcotic pain medications or benzodiazepines do not appear to be a higher risk of discontinuing linaclotide or lubiprostone. Our findings suggest that phenotyping IBS treatment response based on established biological disease mechanisms may be reasonable and feasible, and these findings require validation in future prospective studies. Future efforts to improve phenotyping of IBS treatment response harnessing machine learning techniques to incorporate multi-omics approaches may benefit from the use of a medication persistence endpoint.

Constipation-predominant irritable bowel syndrome (IBS-C) is characterized by abdominal pain associated with symptoms of chronic constipation (such as straining during defecation, lumpy/hard or infrequent stools, sensation of incomplete evacuation during defecation, sensation of anorectal obstruction or blockage during defecation, or the need for manual maneuvers to facilitate defecation). IBS-C is the most expensive and symptomatically bothersome IBS subtype and affects approximately 5% of Americans [ 1 - 3 ]. Several prescription drugs are effective in managing IBS-C [ 4 - 6 ]. As in many chronic medical conditions, IBS-C prescription drug therapy must be taken daily to maintain global relief of symptoms for this incurable illness [ 7 ]. Unlike other diseases in gastroenterology such as inflammatory bowel disease (genetics, imaging, endoscopy, histology) or reflux disease (objective reflux testing), biological patient factors are lacking to optimize IBS treatment selection and treatments are often chosen empirically [ 8 ]. Despite five on-label prescription drug treatments for IBS-C, treatment discontinuation is common, and overall treatment satisfaction in IBS remains poor [ 9 , 10 ]. We recently characterized medication persistence (defined as the length of time between treatment initiation and discontinuation) as an important, valid, and quantifiable outcome measure of IBS treatment response [ 7 ]. By definition, maximizing the likelihood of medication persistence minimizes the risk of treatment discontinuation. This outcome is especially important, because treatment discontinuation typically implies a return of poorly controlled IBS symptoms if alternative therapy is not pursued. As a quantifiable outcome measure, medication persistence facilitates the identification of treatment selection factors which can optimize treatment selection and inform shared decision making on expected treatment outcomes. Visceral hypersensitivity (i.e., alteration in brain–gut interactions) is a core concept and one of the primary mechanisms differentiating IBS-C from functional constipation (FC) [ 11 ]. In practice, fibromyalgia, temporomandibular disorders, chronic fatigue syndrome, migraine headache, osteoarthritis, endometriosis, and chronic pelvic pain/vulvodynia are all associated with pain hypersensitization; these comorbidities were categorized in 2011 by the Trans-NIH Chronic Overlapping Pain Conditions Working Group as Chronic Overlapping Pain Conditions (COPC) [ 12 , 13 ]. Mood disorders such as major depressive disorder and generalized anxiety disorder occur in over 40% of IBS patients, and these disorders are known to influence medication persistence and adherence in chronic health conditions outside of gastroenterology [ 14 , 15 ]. As COPCs and mood disorders occur frequently in IBS populations and are easily identifiable in electronic health records, we performed a cohort study aimed at determining whether COPCs and mood disorders influence medication persistence and treatment discontinuation in IBS-C patients managed with two FDA-approved secretagogues.