Comparative safety of flavocoxid vs prescription NSAIDs among osteoarthritis patients

SUMMARY

Objective

Flavocoxid is a prescription medical food used to manage osteoarthritis (OA) symptoms. Safety concerns based on case reports raised an association with acute liver injury and hypersensitivity pneumonitis. We determined incidence rates (IR) of these safety events in a cohort of new users of flavocoxid and prescription non-steroidal anti-inflammatory drugs (NSAIDs).

Method

MarketScan^® claims data (2006–2017) was used to identify patients initiating flavocoxid or NSAIDs. Propensity score matching (1:2 ratio) was used to balance patient characteristics. Outcomes included hospitalization for hypersensitivity pneumonitis, liver injury, gastrointestinal bleeding, myocardial infarction, and acute kidney injury. Poisson regression was used to calculate IRs and Cox regression for calculating adjusted hazard ratios (aHR).

Results

3,337 flavocoxid and 6,674 NSAID users met eligibility criteria. Before matching, flavocoxid users were older (mean 57 vs 51 years), had more polypharmacy (68% vs 29% taking ≥11 medications). After matching, characteristics were well balanced.

The rate of hypersensitivity pneumonitis was 1.1 (95% CI 0.0–5.9) per 1,000 PY for flavocoxid and 0.0 (95% CI 0.0–2.2) for NSAIDs. For hospitalized liver injury, it was 3.2 (95% CI 0.7–9.3) for flavocoxid and 2.4 (95% CI 0.7–6.1) for NSAIDs, aHR = 1.16, 95% CI 0.23–6.01. A lower rate of GI bleed was observed, IR: 5.3 (1.7–12.3) for flavocoxid and 10.2 (5.9–16.3) for NSAIDs, aHR 0.49 (0.18–1.68). There were no significant differences for MI or AKI.

Conclusion

The rate of hypersensitivity pneumonitis and liver injury associated with flavocoxid was low and minimally elevated compared to NSAIDs. Flavocoxid users had a significantly lower risk for hospitalized GI bleeding. The risk–benefit profile of flavocoxid may warrant reevaluation in light of these findings.

INTRODUCTION

Flavocoxid (Limbrel) is a medical food consisting of two plant-derived flavonoids, baicalin and catechin, compounds shown to have anti-inflammatory properties¹. The ingredients in flavocoxid were classified as Generally Recognized as Safe (GRAS), a regulatory requirement for a medical food. First available in the United States in 2004, approximately 2,000,000 prescriptions were dispensed to an estimated 450,000 patients. Flavocoxid often was prescribed by rheumatologists, orthopedists, and pain management specialists, and commonly prescribed for patients not able to tolerate or otherwise not considered appropriate candidates for prescription non-steroidal anti-inflammatory drugs (NSAIDs). Flavocoxid was available only by prescription and specifically formulated to assist in the management of inflammatory and oxidative processes resulting in joint discomfort and loss of mobility, as is common in osteoarthritis (OA).

The efficacy of flavocoxid was based on two multicenter, randomized, double-blind controlled trials comparing flavocoxid 500 mg twice daily to prescription naproxen 500 mg. The first of these, was a 4-week blinded trial of 105 OA patients². Clinical endpoints were measured via patient- and physician-reported outcomes and included the Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) as well as timed walk measures; results showed comparability between flavocoxid and naproxen. These efficacy results were confirmed in a similarly-designed 12-week second study conducted in 220 participants³.

Between January 2007 and November 2017, the Food and Drug Administration (FDA) received 30 adverse event reports of elevated liver function tests or acute hypersensitivity pneumonitis (HP) associated with the use of flavocoxid products⁴. These events were of varying severity and sometimes occurred shortly after exposure (e.g., within 3 weeks of first use). There were no deaths, and all cases resolved without residual effects after discontinuation. The FDA subsequently requested that the manufacturer recall flavocoxid due to these safety concerns, a recall that occurred on 12/21/2017.

Despite the importance of these potential safety signals, the public health concern is not easily contextualized without an understanding of the absolute incidence of liver injury and HP, and comparative risk vs alternative treatments. Flavocoxid typically was used to treat OA-associated symptoms, so comparison with therapies commonly used to treat OA such as NSAIDs seems reasonable. NSAIDs have been associated with liver injury and may cause adverse effects including gastrointestinal (GI) bleeding, unfavorable changes in cardiovascular risk factors (e.g., hypertension) leading to increased risk for myocardial infarction (MI), and acute kidney injury (AKI). The purpose of this analysis was to determine the incidence rates (IR) of HP and severe acute liver injury associated with flavocoxid and NSAID use, with additional information regarding GI bleeding, MI, and AKI provided for risk contextualization.

METHODS

Cohort eligibility and exposure

We queried the IBM^® MarketScan^® Research claims database (2006–2017), a data source containing administrative records contributed by large employers, managed care organizations, pharmacies, health care systems and hospitals to identify patients age ≥18 with a history of one of twelve common flavocoxid indications classified based on the International Classification of Diseases (ICD) code categories⁵ (Appendix Table I) who initiated flavocoxid or oral prescription NSAIDs. Data are collected for classification and billing purposes, and generally complete in capturing all compensable medical services submitted for reimbursement by insurance. Over-the-counter(OTC) medications, supplements not dispensed by prescription, or healthcare services not reimbursed by insurance (e.g., samples) are generally not captured. We required a baseline of ≥12 months with continuous enrollment (medical and pharmacy) for covariate/confounder assessment and to enable a new user design⁶. Date of first filled flavocoxid or oral NSAID prescription defined the ‘index date’; the new user design allowed no prior prescriptions for either in the preceding 6 months. Patients were excluded if they had diagnosis codes for malignancy, ignoring non-melanoma skin cancer, skilled nursing facility care (to select for community-dwelling individuals), hospice, or prior history of drug induced liver injury (DILI) or HP using all available preceding data (if more than the 12 month baseline was available). Flavocoxid and NSAID exposure was defined based on the days’ supply in the pharmacy data. Exposure was analyzed as-treated, based on consecutive prescriptions; each prescription fill allowed for a 30 day extension to allow for less-than-perfect adherence and was truncated by a new refill. A sensitivity analysis allowed for a ninety-day extension.

Outcomes of interest

The two primary outcomes were those mentioned in the FDA case reports: hospitalization for liver injury using both a more specific (i.e., drug related) and sensitive definition, and hypersensitivity pneumonitis (ICD-9 code 495.9, HP due to unspecified cause and its corresponding ICD-10 code J67.9). These outcomes were identified in physician diagnosis codes on outpatient encounters (HP) or hospital discharge diagnosis codes(all others). Liver injury algorithms were validated in the FDA Sentinel system⁷. While these had wide variability in their positive predictive values (PPV), given our focus on absolute IR, we followed guidance to, “identify all possible events and utilize the ICD-9-CM codes employed … to screen for potential events”. The HP definition has been used in prior studies with high PPV(95%) and was adapted to capture acute events with high sensitivity, requiring only one physician diagnosis code⁸. Additional validated outcomes included hospitalized GI bleed (using a sensitive, and a more specific definition)⁹, acute myocardial infarction (MI)¹⁰, acute kidney injury (AKI)¹¹(Appendix Table I).

Statistical analysis

To balance characteristics between flavocoxid and prescription NSAID exposure groups, patient were grouped into strata with same age, sex, indication for flavocoxid, and calendar year of initiation. Propensity scores (PS) were used to balance patient characteristics using an optimal matching approach within each strata. Two PS matching approaches were used: 1:2 fixed ratio, to best achieve covariate balance and show Table I comparisons, and 1:10 variable ratio matching to maximize sample size and precision of effect estimates. Factors in the PS were selected based on clinical interest and hypothesized exposure with the outcome: demographics (age; sex, rural/urban residence), comorbidity groups as classified by the Healthcare Utilization Project Clinical Classification Software (CCS) (a taxonomy that groups ICD-9/10 codes for >200 health conditions for diagnosis, and >200 procedure categories)⁵ medication ATCs, and healthcare utilization (recent hospitalization, number of physician visits, total number of unique medication classes). Covariates in the logistic regression used to estimate the PS had prevalence≥1% and p-value ≤0.15 in univariate analysis for exposure. Standardized mean differences (SMD) were used to assess covariate balance, with imbalance defined as SMD≥0.10¹².

Table I.

Patient characteristics of pre- and post-matched^* flavocoxid and NSAID users

Empty Cell	Pre- Match			Post 1:2 Fixed Ratio Propensity Score Matching
Baseline Characteristics	Flavocoxid	NSAIDs	SMD	Flavocoxid	NSAIDs	SMD
N	4,145	975,965		3,337	6,674
Age in Years, Mean (SD)	57.4 (13.1)	50.6 (10.2)	0.59	56.5 (12.6)	56.5 (12.6)	0.00
Female, %	71.1	66.6	0.10	70.3	70.3	0.00
Health conditions associated with flavocoxid use †
Rheumatoid arthritis (CCS 202)	11.4	0.2	0.49	5.8	5.8	0.00
Osteoarthritis (CCS 203)	46.2	10.2	0.87	40.5	40.5	0.00
Other joint disorders (CCS 204)	60.7	33.9	0.56	58.3	58.3	0.00
Spondylosis (CCS 205)	63.5	57.8	0.12	61.3	61.3	0.00
Osteoporosis (CCS 206)	7.2	0.2	0.37	3.5	3.5	0.00
Foot deformities (CCS 208)	10.1	0.7	0.43	7.1	7.1	0.00
Other deformities (CCS 209)	4.8	0.1	0.31	1.9	1.9	0.00
Systemic lupus (CCS 210)	5.9	0.1	0.35	2.1	2.1	0.00
Other conn. tissue d/o (CCS 211)	67.0	55.7	0.23	64.9	64.9	0.00
Other bone/msk CCS 212	19.5	3.4	0.53	14.0	14.0	0.00
Joint disorder - traumatic (CCS 225)	14.7	1.5	0.50	9.5	9.5	0.00
Gout (CCS 54)	3.2	0.1	0.24	1.8	1.8	0.00
Selected covariates ‡
Sprains and strains (CCS Diagnosis 232)	27.2	18.9	0.20	25.7	23.3	0.06
Hyperplasia (CCS Diagnosis 164)	5.4	2.5	0.15	4.4	5.6	0.06
Upper gastrointestinal endoscopy (CCS Procedure 70)	4.3	1.5	0.16	3.9	3.1	0.04
Other non-OR therapeutic on skin (CCS Procedure 174)	2.5	1.0	0.11	2.5	1.9	0.04
Duloxetine use (ATC N06AX21)	13.0	3.5	0.35	10.4	7.3	0.11
Alendronate use (ATC M05BA04)	5.8	1.7	0.22	4.4	6.3	0.09
Number of ATC medication classes, %
0–5	9.0	34.3	0.65	11.4	10.6	0.04
6–10	22.7	36.5	0.31	25.7	27.1	0.01
≥11	68.3	29.1	0.85	63.0	62.3	0.03
Selected Medications, (%)
Anticoagulants	10.4	4.2	0.24	9.3	9.5	0.01
Proton pump inhibitors	43.3	18.8	0.55	37.2	35.8	0.03
Antihypertension	58.6	35.4	0.48	55.0	57.2	0.04
Antidiabetics	13.8	9.6	0.13	13.5	13.9	0.01
Beta blockers	26.2	13.8	0.31	24.2	26.1	0.04
Statins	38.0	21.4	0.37	37.4	37.4	0.00
Lipids	12.9	5.5	0.26	11.6	11.9	0.01
Antibiotics	55.5	45.1	0.21	49.5	50.3	0.02
Steroid use						0.07
None	77.6	85.5	0.21	82.6	81.3
<7.5 mg/day	21.1	13.9	0.19	16.7	17.5
≥7.5 mg/day	1.3	0.5	0.08	0.7	1.3
Hospitalization, %	7.4	6.9	0.02	6.2	7.0	0.03
Number of physician visits, %						0.03
0–5	61.3	87.1	0.62	55.1	53.6
6–10	29.2	11.0	0.47	32.7	34.0
≥11	9.5	2.0	0.33	12.2	12.4

SMD = standardized mean difference. SMDs >0.10 generally indicate important imbalance.

CCS = AHRQ Clinical Classification Software comorbidity list.

CTD = connective tissue disease; d/o = disorder.

^*flavocoxid users were matched 1:2 to prescription NSAID users using fixed ratio optimal matching.

^†as defined by the AHRQ CCS for diagnosis.

^‡abbreviated list of the more complete set of AHRQ CCS and ATC medication categories included in the propensity score; the several most imbalanced after matching were shown.

Poisson regression was used to calculate IR per 1000 person-years (PY); exact methods were used for 95% confidence intervals (CI) for incidence rate ratios and incidence rate differences. Cox regression was used for calculating adjusted hazard ratios (aHR), controlling for any covariates that had residual imbalances based on SMD. Patients were censored when they discontinued flavocoxid or NSAID, experienced the outcome of interest, lost coverage, or on 12/31/2017.

All analyses were conducted in SAS^® 9.4. The university institutional review board approved the study. Results were reported according to Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines¹³.

Results

After applying eligibility criteria (Appendix Figure), 3,337 new flavocoxid users were matched to 6,674 new prescription NSAID users (fixed 1:2 ratio analysis). Prior to matching, many imbalances between flavocoxid and prescription NSAID users were observed (Table I), and flavocoxid users were generally older and sicker in all respects (e.g., prevalence of comorbidities, polypharmacy, healthcare utilization). All imbalances were greatly attenuated after PS matching, with only concomitant duloxetine use having a residual SMD>0.10.

The median (interquartile range (IQR)) duration of exposure to flavocoxid and NSAIDs was short (median (IQR) 61(61, 114) days for flavocoxid, 61(61, 85) days for NSAIDS; in the sensitivity analysis, median (IQR) follow-up duration was 121(121,181) and 121(121,166) days. As shown in Table II, and based on 1 flavocoxid-associated case (1:2 fixed ratio analysis), the IR for HP was 1.1 (95% CI 0.0–5.9) per 1,000 PY; there were no NSAID cases. The incidence rate for hospitalized liver injury was numerically greater in flavocoxid vs NSAID users (3.4 vs 1.3/1000 PY) but not significant(aHR = 2.40, 95% CI 0.63–9.11). Rates of DILI were 1.1/1000 PY and similar between flavocoxid and NSAID users.

Table II.

Incidence rates and adjusted hazard ratios of safety outcomes in flavocoxid and NSAID users

Outcome	Match			Exposure	Events, n	Person year	IR per 1000 PY (95% CI)	Adjusted* HR (95% CI)	IRD per 1000 PY (95% CI)
Empty Cell	Fixed or Variable ratio	Ratio	Caliper width
Hypersensitivity pneumonitis	Fixed	2	0.05	Flavocoxid	1	945.8	1.1 (0.0–5.9)	NA	1.1 (−1.0, 3.1)
				NSAID	0	1677.5	0.0 (0.0–2.2)
	Variable	10	0.01	Flavocoxid	1	881.4	1.1 (0.0–6.3)	3.88 (0.24–63.51)	1.0 (−1.2, 3.2)
				NSAID	1	6105.1	0.2 (0.0–0.9)
Any hospitalized liver injury	Fixed	2	0.05	Flavocoxid	3	944.8	3.2 (0.7–9.3)	1.16 (0.23–6.01)	0.8 (−3.5, 5.1)
				NSAID	4	1674.5	2.4 (0.7–6.1)
	Variable	10	0.01	Flavocoxid	3	880.4	3.4 (0.7–10.0)	2.40 (0.63–9.11)	2.1 (−1.9, 6.0)
				NSAID	8	6101.9	1.3 (0.6–2.6)
Hospitalized drug induced liver injury	Fixed	2	0.05	Flavocoxid	1	946.0	1.1 (0.1–5.9)	0.53 (0.05–6.33)	−0.7 (−3.6, 2.2)
				NSAID	3	1675.4	1.8 (0.4–5.2)
	Variable	10	0.01	Flavocoxid	1	881.6	1.1 (0.0–6.3)	1.05 (0.13–8.73)	0.2 (−2.2, 2.5)
				NSAID	6	6102.8	1.0 (0.4–2.1)
Hospitalized GI bleeding (sensitive)	Fixed	2	0.05	Flavocoxid	5	945.3	5.3 (1.7–12.3)	0.49 (0.18–1.34)	−4.8 (−11.6, 1.8)
				NSAID	17	1673.3	10.2 (5.9–16.3)
	Variable	10	0.01	Flavocoxid	3	881.4	3.4 (0.7–9.9)	0.55 (0.18–1.68)	−4.6† (−9.0, −0.2)
				NSAID	49	6095.7	8.0 (5.9–10.6)
Hospitalized GI bleeding (specific)	Fixed	2	0.05	Flavocoxid	1	946.1	1.1 (0.0–5.9)	0.30 (0.03–2.93)	−1.9 (−5.3, 1.4)
				NSAID	5	1676.1	3.0 (1.0–7.0)
	Variable	10	0.01	Flavocoxid	1	881.7	1.1 (0.0–6.3)	0.36 (0.05–2.70)	−1.7 (−4.2, 0.9)
				NSAID	17	6102.3	2.8 (1.6–4.5)
Myocardial infarction	Fixed	2	0.05	Flavocoxid	2	945.7	2.1 (0.3–7.6)	0.91 (0.15–5.63)	−0.3 (−4.0, 3.5)
				NSAID	4	1677.0	2.4 (0.6–6.1)
	Variable	10	0.01	Flavocoxid	1	881.4	1.1 (0.0–6.3)	0.30 (0.03–2.92)	−3.5† (−6.2, −0.7)
				NSAID	28	6099.6	4.6 (3.1–6.6)
Acute kidney injury	Fixed	2	0.05	Flavocoxid	13	943.8	13.8 (7.3–23.6)	1.43 (0.70–2.93)	3.0 (−6.0, 12.0)
				NSAID	18	1674.6	10.7 (6.4–17.0)
	Variable	10	0.01	Flavocoxid	10	880.7	11.4 (5.4–20.9)	1.49 (0.70–3.20)	3.0 (−4.4, 10.4)
				NSAID	51	6095.9	8.4 (6.2–11.0)

CI = Confidence interval, GI = Gastrointestinal, HR = Hazard Ratio; IRD=Incidence rate difference; NA = Not able to estimate given 0 cases in referent group; NSAIDs = Non-steroidal anti-inflammatory drugs, PS = propensity score; PY = person years; NA = not assessable (zero events in referent exposure group).

^*Reference = NSAID, adjusted for covariates with SMD>0.1 after propensity score match.

^†denotes an incidence rate difference with a 95% confidence interval that does not include 0.

A numerically lower rate of GI bleeding was observed for flavocoxid (IR:5.3) vs NSAIDs: 10.2, aHR = 0.49, 95% CI 0.18–1.34). The incidence rate difference resulting from the variable ratio GI bleeding analysis was significant (incidence rate difference −4.6, 95% CI −9.0,−0.2). There were no significant differences between for MI or AKI, although the confidence interval of the MI incidence rate difference excluded 0 (−3.5, 95% CI −6.2,−0.7) favoring flavocoxid. Results were robust in sensitivity analyses(Appendix Table II).

Discussion

In this population-based analysis, we found the incidence rate for HP and hospitalized liver injury associated with flavocoxid was low (incidence rate 1–3/1000 PY) and the excess risk small (incidence rate difference 1–2 per 1,000 PY) compared to prescription NSAIDs. For risk contextualization, the rate of hospitalized GI bleeding was numerically higher in NSAID users, a rate difference of approximately 2–5 per 1,000 PY. Risks for MI and AKI were similar between flavocoxid and NSAID users. In total, and assuming that HP, liver injury, and GI bleeding represent events of relatively comparable severity, the safety profile of flavocoxid seems to be on par with prescription NSAID use with respect to these outcomes.

The safety events were identified based on claims-based coding algorithms used in prior studies, many validated against medical record review. For HP and liver injury outcomes, we purposefully increased the sensitivity of our case-finding approach by requiring only one diagnosis code recognizing that not all events identified approach would be true cases. Thus, our estimates likely reflect the upper bound of the absolute risk of these two events of such concern to FDA. For other outcomes, the comparability of our absolute IR to those observed in a clinical trial with adjudication gives greater weight to the generalizability of our findings. For example, the PRECISON trial¹⁴ compared the safety between three prescription NSAIDs. The rate of serious gastrointestinal events was approximately 8.5 (naproxen) and 8.8 (ibuprofen) per 1000 PY, similar to the rate of 8–10 observed in this study. Likewise, the rate of nonfatal MI in PRECISON was 3.9 (naproxen) and 5.8 (ibuprofen) per 1,000 PY, similar to our rate of 2–5.

Strengths of our study include use of a large cohort with a new user, active comparator design that compared rigorously-constructed, well-balanced, propensity score-matched cohorts. While potential for confounding exists related to unmeasured (e.g., race), misclassified, or time-varying factors that changed shortly after matching, flavocoxid users generally were older, sicker, and with greater multimorbidity. Thus, any residual confounding would be expected to bias in favor of the NSAID group. We recognize patients may have used over-the-counter NSAIDs not captured in this data; it is unknown whether flavocoxid patients might be more, less or similarly likely to take OTC NSAIDs compared to prescription NSAID users. If flavocoxid users were as or more likely to take OTC NSAIDs compared to prescription NSAID users, then results for NSAID-associated outcomes would be biased towards the null. However, this would not be expected to influence the HP outcome, given no known association with NSAIDs. Finally, while exposure to both flavocoxid and prescription NSAIDs was short (median 2 months), this duration of exposure nevertheless is relevant given the short exposure-outcome associations described by FDA4. Indeed, most HP and DILI cases occurred shortly after initial usage (1–3 weeks for HP, 2–3 months for DILI), and resolved upon discontinuation¹⁵. Moreover, it reflects the real-world experience of some patients needing these treatments only episodically, and informs the safety profile of this use pattern.

In conclusion, we found that absolute risk of HP and liver injury associated with flavocoxid was low and increased over prescription NSAIDs by 1–2/1000 PY. This difference was similar in magnitude but in the opposite direction for hospitalized GI bleeding. Higher risk patients and those with a greater burden of multimorbidity were more likely to have been prescribed flavocoxid rather than prescription NSAIDs. Given these results and the previously-demonstrated efficacy of flavocoxid to alleviate physical discomfort, improve mobility and reduce pain, the risk–benefit profile of flavocoxid may warrant re-evaluation in light of its overall efficacy and safety profile.

Supplementary Material

Appendix Table 1

Appendix Table. I Algorithms to Identify Outcomes of Interest

Appendix Table 2

Appendix Table. II (sensitivity analysis): Incidence Rates and Adjusted Hazard Ratios of Safety Outcomes in Flavocoxid and NSAID Users using 90-day extension

Appendix Figure

Appendix Fig. Attrition Table Describing Cohort Selection