The effect of drug interactions on bleeding risk associated with warfarin therapy in hospitalized patients

Abstract

Background. Bleeding is a serious adverse drug reaction associated with warfarin therapy, often induced by interacting co-medication.

Methods. We investigated the frequency and clinical consequences of warfarin drug interactions utilizing medical records of 6,772 warfarin-treated in-patients of Turku University Hospital.

Results. A total of 48% of warfarin-treated in-patients were exposed to interacting co-medication. Adjusted odds ratio (OR) for bleeding was highest for cytochrome P450 2C9 (CYP2C9) inhibitors (OR 3.6; 95% confidence interval (CI) 2.4–5.6). Non-selective non-steroidal anti-inflammatory drugs (NSAID) and coxibs were associated with a bleeding risk of a similar magnitude (OR 2.6; 95% CI 1.6–4.2 and OR 3.1; 95% CI 1.4–6.7, respectively). Selective serotonin re-uptake inhibitors (SSRI) were associated with a remarkably higher bleeding risk than non-SSRIs (OR 2.6; 95% CI 1.5–4.3 and OR 1.2; 95% CI 0.3–4.3, respectively). Odds ratio for bleeding in the platelet aggregation inhibitor group was 1.6 (95% CI 0.8–3.1).

Conclusion. We conclude that co-medication in warfarin-treated in-patients is common and should be carefully evaluated to decrease the bleeding risk associated with warfarin therapy.

• Antidepressants
• coxibs
• CYP2C9 inhibitors
• drug interactions
• hemorrhage
• NSAIDs
• platelet aggregation inhibitors
• warfarin

Introduction

Warfarin is a widely prescribed oral anticoagulant for thromboembolic therapy in Europe and in North America. Warfarin interferes with vitamin K metabolism by inhibiting vitamin K epoxide reductase and, thus, synthesis of biologically active coagulation factors [1]. Warfarin is administered as a racemic mixture of R- and S-enantiomers, the S form being three to five times more potent anticoagulant than the R form. The therapeutic index of warfarin is low, and therefore the anticoagulation response is routinely measured by the international normalized ratio (INR). Nevertheless, over-anticoagulation and bleeding often complicate warfarin therapy. Patients receiving oral anticoagulant therapy have approximately a 3-fold risk of upper gastrointestinal bleeding (UGIB) compared to non-users [2–4]. Although bleeding can occur at the therapeutic level, the risk increases with the increasing intensity of anticoagulation. In clinical practice, however, achieving this safe yet effective range is often a challenge because of large interindividual variation in dose response. This variation is attributable to both genetic and environmental factors. Contrary to the influence of genetic factors, the effect of environmental factors on individual anticoagulant response is unstable. Factors such as dietary intake of vitamin K, liver diseases, and co-medications cause fluctuation to the anticoagulant response and a risk for bleeding.

Warfarin drug interactions have been shown to increase the bleeding risk [5–7]. Drugs that inhibit cytochrome P450 2C9 isoenzyme (CYP2C9) enzyme prevent the oxidative metabolism of S-warfarin, the more potent enantiomer of racemic warfarin, increasing its plasma concentration and leading to enhanced anticoagulation. Moreover, the bleeding risk associated with warfarin therapy may also be raised by drugs that impair platelet function or cause gastroerosion. Concurrent non-steroidal anti-inflammatory drug (NSAID) or platelet aggregation inhibitor medication with warfarin has been reported to increase the risk for upper gastrointestinal bleeding (UGIB) close to 2- or 3-fold, respectively [8], [9], compared to warfarin alone. However, systematic data on the safety of warfarin treatment with selective serotonin re-uptake inhibitors (SSRI) and CYP2C9 inhibitors are scarce or absent [10]. The effect of oral glucocorticoids on the UGIB risk remains a controversial issue in the medical literature [11–13], but according to one study combining corticosteroid therapy to anticoagulation with warfarin caused almost an 8-fold increase in the UGIB rate [14]. Instead, proton pump inhibitors (PPI) have been shown to reduce the UGIB risk in patients taking traditional NSAIDs and acetylsalicylic acid, and to prevent recurrence of UGIB in patients on warfarin [15], [16]. In contrast to interactions changing the exposure to warfarin, the interactions affecting the platelet function may be largely undetectable by INR monitoring, which complicates their detection and preventability in clinical practice. The aim of this register study was to investigate the frequency and the clinical consequences of drug-–drug interactions between warfarin and co-medication that has potential to increase the risk for bleeding in a hospital setting.

Key messages

• Concomitant use of cytochrome P450 2C9 (CYP2C9) inhibitors was associated with the highest increase of bleeding risk, and, especially, these risks could have been avoided by an alternative choice of medication.

• In warfarinized patients, coxibs were associated with a similar risk for bleeding as non-selective NSAIDs, but antidepressants not inhibiting re-uptake of serotonin were associated with a significantly lower risk for bleeding compared with SSRIs.

• Co-medication with warfarin should be carefully evaluated to decrease the risk for bleeding.

Materials and methods

Setting

The data were obtained from the medical record databases of Turku University Hospital, Turku, Finland [17]. Exposures to warfarin with or without concomitant drugs associated with increased risk for bleeding were extracted from a medication database by utilizing the Anatomical Therapeutic Chemical (ATC) codes. The study period spanned 8.5 years (from 1 July 1996 to 31 December 2004), and at the time there were no programs available for checking potential drug–drug interactions. The medication data contained information on the name, strength, dosage form, and dosage of the drug, also the starting and stopping dates of the medication and ward, and the patient's social security number to identify the users of the requested medications. Thereafter, the concomitant use or non-use of interacting medication with warfarin was analyzed for each patient by the use of the patient's social security number and the starting and stopping dates of the medications. The data were reviewed manually in order to ascertain exposure to warfarin with or without concomitant interacting medication. Subsequent to identification of study cases (by social security number), the bleeding diagnoses and the laboratory values describing the clinical effects of interactions were searched in the ICD-10 (International Classification of Diseases, 10th revision) coded diagnose database and in the laboratory database. All the searches were done after approval of the study protocol by the local authorities responsible for the hospital registries.

Abbreviations

Table

ATC	anatomical therapeutic chemical
B-ERYT	BLOOD ERYTHROCYTE COUNT
B-Throm	platelet count
CI	confidence interval
CYP2C9	cytochrome P450 2C9 isoenzyme
Hb	hemoglobin
HCT	hematocrit
ICD-10	International Classification of Diseases, 10th revision
INR	international normalized ratio
MCH	mean cellular hemoglobin
MCV	mean corpuscular volume
non-SSRI	non-selective serotonin re-uptake inhibitor.
NSAID	non-steroidal anti-inflammatory drug
OR	odds ratio
PPI	proton pump inhibitor
SSRI	selective serotonin re-uptake inhibitor
UGIB	upper gastrointestinal bleeding

Study population and outcome measures

The study population comprised 8,615 warfarin treatment periods (6,772 warfarin-treated patients) in the internal medicine (5,584 treatment periods; 64.8% of total), pulmonary medicine (875; 10.2%), neurology (1,366; 15.9%), and oncology wards (790; 9.2%) at the Turku University Hospital during the study period. A treatment period is defined as each recorded new admission to the hospital. The personnel of the wards were not aware of the study. The study population was divided into seven groups: patients receiving warfarin concomitantly with platelet aggregation inhibitors; antidepressants, either non-SSRI or SSRI; NSAID, either non-selective NSAIDs or coxibs; or CYP2C9 inhibitors (i.e. interaction groups); and those receiving warfarin only (i.e. control group). Thus, a patient on a regimen of several interacting drugs could be included in more than one group. This was taken into account in the statistical analysis. The selected interaction cases fulfilled the criterion of at least 7 days co-administration of warfarin and interacting drug—except for CYP2C9 inhibitors, for which concomitant use of 2 days was considered sufficient to cause interaction [9]. The possible use of PPI (esomeprazole, lansoprazole, omeprazole, pantoprazole, or rabeprazole) and oral glucocorticoid (dexamethasone, hydrocortisone, methylprednisolone, prednisolone, or prednisone) medications were also reviewed from the medication records, and the patients were further divided into users and non-users of these medications accordingly. All interacting drugs are listed in Table I.

Table I.  List of interacting drugs.

Platelet aggregation inhibitor	Frequency, n (%)
Acetylsalicylic acid	334 (71.4)
Dipyridamole	96 (20.5)
Clopidogrel	34 (7.3)
Ticlodipine	4 (0.9)
Total	468 (100)
No interactions were found for iloprost, combinations (acetylsalicylic acid and dipyridamole), abciximab, eptifibatide, and tirofiban.
SSRI	Frequency, n (%)
Citalopram	677 (78.5)
Fluoxetine	105 (12.2)
Sertraline	40 (4.6)
Paroxetine	33 (3.8)
Clomipramine	8 (0.9)
Total	863 (100)
No interactions were found for escitalopram.
SSRI = selective serotonin re-uptake inhibitor.
Non-SSRI	Frequency, n (%)
Mirtazapine	184 (90.6)
Moclobemide	19 (9.4)
Total	203 (100)
No interactions were found for reboxetine.
Non-SSRI = non-selective serotonin re-uptake inhibitor.
Non-selective NSAID	Frequency, n (%)
Ibuprofen	355 (27.9)
Diclofenac	216 (17.0)
Naproxen	201 (15.8)
Ketoprofen	164 (12.9)
Nimesulide	143 (11.2)
Indometacin	67 (5.3)
Meloxicam	50 (3.9)
Ketorolac	30 (2.4)
Nabumetone	13 (1.0)
Tolfenamic acid	13 (1.0)
Piroxicam	8 (0.6)
Etodolac	4 (0.3)
Tenoxicam	3 (0.2)
Tiaprofenic acid	3 (0.2)
Aceclofenac	2 (0.2)
Mefenamic acid	1 (0.1)
Total	1273 (100)
No interactions were found for lornoxicam, dexibuprofen, and dexketoprofen.
NSAID = non-steroidal anti-inflammatory drug.
Coxib	Frequency, n (%)
Rofecoxib	121 (49.2)
Celecoxib	105 (42.7)
Etoricoxib	17 (6.9)
Valdecoxib	3 (1.2)
Total	246 (100)
No interactions were found for parecoxib.
CYP2C9 inhibitor	Frequency, n (%)
Metronidazole	792 (41.1)
Fluconazole	435 (22.6)
Amiodarone	343 (17.8)
Phenytoin	170 (8.8)
Miconazole	55 (2.9)
Sulfamethoxazole	48 (2.5)
Tamoxifen	35 (1.8)
Zafirlukast	24 (1.2)
Gemfibrozil	17 (0.9)
Fluvoxamine	10 (0.5)
Total	1929 (100)

No interactions were found for voriconazole.

CYP2C9 = cytochrome P450 2C9 isoenzyme.

For evaluating the clinical effects of drug interactions the ICD-10-coded bleeding diagnoses were reviewed over the period of co-administration (or warfarin administration in the control group) using the unique social security number for patient identification. Diagnoses were divided into four categories for further analyses: all, intracranial, and upper and lower gastrointestinal bleeding diagnoses. In addition, hematocrit (HCT), mean and minimum hemoglobin concentrations (Hb and Min Hb, respectively), blood erythrocyte count (B-Eryt), mean cellular volume and hemoglobin (MCV and MCH, respectively), platelet count (B-Throm), and INR from each treatment period were examined. These laboratory values were collected from the second day after the beginning of the interaction or the warfarin only treatment to the fifth day after the end of the exposure. Means of these values were calculated in case of multiple measurements. INR values greater than 7.0 were replaced with the value 7.0, since the laboratory did not report the exact values exceeding 7.0.

Statistical analysis

Sex distribution and the proportions of treatment periods with concomitant exposure to PPI or oral glucocorticoid medication were compared between the interaction and control groups using binary logistic regression analysis. The correlation between the treatment periods measured from the same patient was taken into account by using generalized estimation equations. Mean ages between the groups were compared with the linear mixed model with patients as random effect, which takes into account the correlation between the treatment periods measured from the same patient. The differences in the risk for bleeding and in the risk of being outside the Hb target range (<117 g/L for women and <134 g/L for men), platelet count (>360×10⁹/L) and INR value (>3.0) between the interaction and control groups were tested with binary logistic regression using generalized estimation equations [18]. Analyses were adjusted for age, sex, and study ward, and PPI and oral glucocorticoid medication. The control group was used as a reference group in logistic models. The results from logistic models were quantified by using odds ratios (OR) with their 95% confidence intervals (CI). The linear mixed model with patients as random effect after adjustment for the factors given above was used in the comparison of the means of the laboratory variables between the interaction and control groups [19]. Dunnett's method with controls as a comparison group was used for multiple comparisons [20]. P-values less than 0.05 were considered statistically significant. Statistical analyses were done using SAS System for Windows, release 9.1 (SAS Institute Inc., Cary, NC).

Results

Interaction frequencies

Altogether, there were 8,615 warfarin treatment periods among all 158,217 treatment periods in the chosen wards during the 8.5 years studied. In 1,929 treatment periods (22.4% of all warfarin treatment periods) warfarin was used concomitantly with a CYP2C9 inhibitor, in 1,273 (14.8%) and in 246 (2.9%) treatment periods with non-selective NSAID and coxibs, respectively. SSRIs and non-SSRIs were co-administered in 863 (10.0%) and in 203 (2.4%) warfarin treatment periods, respectively. Co-administration of a platelet aggregation inhibitor was found in 468 (5.4%) warfarin treatment periods. Altogether, 3,221 (47.6%) warfarin-treated patients were exposed to at least one interacting drug. The most common interacting drugs in each drug category were metronidazole (41.1% of treatment periods with CYP2C9 inhibitors) (Table I), ibuprofen (27.9% of treatment periods with non-selective NSAIDs), rofecoxib (49.2% of treatment periods with coxibs), citalopram (78.5% of treatment periods with SSRIs), mirtazapine (90.6% of treatment periods with non-selective SSRIs), and acetylsalicylic acid (71.4% of treatment periods with platelet aggregation inhibitors). The frequencies of all interaction treatment periods are shown in Table I.

A statistically significant difference was observed in sex distribution of all interaction groups except the CYP2C9 inhibitor group, when compared with controls (Table II). Similarly, in all interaction groups patients’ age was also slightly higher than in the control group. The frequency of treatment periods with concomitant exposure to PPI or oral glucocorticoid medication was significantly higher in all interaction groups than in the control group (Table II).

Table II.  The demographic characteristics of cases and controls.

	Platelet aggregation inhibitor	Antidepressant		NSAID
		SSRI	Non-SSRI	Non-selective	Coxib	CYP2C9 inhibitor	Control
Treatment periods, n (% of total)	468 (5.4)	863 (10.0)	203 (2.4)	1273 (14.8)	246 (2.9)	1929 (22.4)	3633 (42.2)
Patients, n (% of total)	396 (5.8)	828 (12.2)	199 (2.9)	1067 (15.8)	222 (3.3)	1407 (20.8)	3614 (53.4)
Sex, male/female	279/189^a	406/457^b	94/109^a	621/652^b	111/135^a	1051/878	1964/1669
Mean age, years (range)	69^a (22–94)	69^b (15–96)	67^b (15–87)	68^b (17–97)	68^b (24–92)	67^b (18–99)	66 (14–97)
Proportion (%) of treatment periods with concomitant exposure to PPI	25.9 ^b	36.7 ^b	43.4 ^b	27.7 ^b	42.3 ^b	38.5 ^b	13.8
Proportion (%) of treatment periods with concomitant exposure to oral glucocorticoid	18.8^b	26.7^b	32.5^b	43.4^b	44.7^b	47.2^b	1.5

SSRI = selective serotonin re-uptake inhibitor; Non-SSRI = non-selective serotonin re-uptake inhibitor; NSAID = non-steroidal anti-inflammatory drug; CYP2C9 = cytochrome P450 2C9 isoenzyme; PPI = proton pump inhibitor.

^aP<0.05 for comparison to the control.

^bP<0.001 for comparison to the control.

Clinical outcome of the interactions

A total of 265 warfarin treatment periods (3.1% of all warfarin treatment periods) were linked to at least one bleeding diagnosis reported over the period of co-administration, or warfarin administration in the control group. The highest percentage (5.4%) of treatment periods with bleeding diagnoses was found in the CYP2C9 inhibitor group (Table III). The corresponding figure in the control group was 1.5%. The non-SSRI group was the only group with no upper or lower gastrointestinal bleeding diagnoses. The number of intracranial bleedings was low in all groups.

Table III.  The frequencies of warfarin treatment periods linked with a bleeding diagnosis and the bleeding risks caused by co-medications.

	All bleeding diagnoses			Upper GI bleeding diagnoses			Lower GI bleeding diagnoses			Intracranial bleeding diagnoses
	n (% of total)	OR	(95% CI)	n (% of total)	OR	(95% CI)	n (% of total)	OR	(95% CI)	n (% of total)	OR	(95% CI)
Platelet aggregation inhibitor	13 (2.8)	1.62	(0.84–3.13)	3 (0.6)	2.18	(0.56–8.45)	1 (0.2)	0.83	(0.10–7.05)	0 (0)
Antidepressant
SSRI	35 (4.1)	2.57^b	(1.54–4.28)	13 (1.5)	5.49^a	(1.79–16.85)	3 (0.3)	1.33	(0.33–5.36)	3 (0.3)	1.47	(0.35–6.17)
Non–SSRI	4 (2.0)	1.15	(0.31–4.25)	0 (0)			0 (0)			1 (0.0)	2.12	(0.24–18.69)
NSAID
Non-selective	44 (3.5)	2.57^b	(1.56–4.23)	8 (0.6)	2.79	(0.88–8.91)	10 (0.8)	3.96 ^a	(1.13–13.93)	5 (0.4)	1.68	(0.47–6.02)
Coxib	12 (4.9)	3.10^b	(1.44–6.67)	2 (0.8)	2.91	(0.33–25.79)	1 (0.4)	1.73	(0.24–12.51)			0 (0)
CYP2C9 inhibitor	104 (5.4)	3.64^b	(2.36–5.62)	29 (1.5)	5.71 ^b	(2.17–15.01)	13 (0.7)	3.12 ^a	(1.04–9.28)	11 (0.6)	2.39	(0.64–8.86)
Control	53 (1.5)			8 (0.2)			9 (0.2)					8 (0.2)

Multivariate logistic regression, adjusted for age, sex, ward, proton pump inhibitor and oral glucocorticoid medications. Adjustment for ward was impossible in lower GI and intracranial bleeding categories because the number of diagnoses in some specialities was zero.

OR = odds ratio; CI = confidence interval; GI = gastrointestinal; SSRI = selective serotonin re-uptake inhibitor; non-SSRI = non-selective serotonin re-uptake inhibitor; NSAID = non-steroidal anti-inflammatory drug; CYP2C9 = cytochrome P450 2C9 isoenzyme.

^aP<0.05 for comparison to the control.

^bP<0.001 for comparison to the control.

A significantly increased risk for bleeding was associated with all but the platelet aggregation inhibitor and the non-SSRI groups when compared to the controls. The greatest risk was seen in the CYP2C9 inhibitor group (OR 3.64, 95% CI 2.36–5.62) (Table III). Co-administration of warfarin with coxibs was associated with an increased bleeding risk of a similar magnitude as with non-selective NSAIDs (OR 3.10, 95% CI 1.44–6.67, and OR 2.57, 95% CI 1.56–4.23, respectively); P=0.62 for difference. In the SSRI group, the risk for bleeding was 2.57-fold compared to controls (95% CI 1.54–4.28). Concomitant CYP2C9 inhibitor and SSRI medication with warfarin was associated with greatly increased risk for upper gastrointestinal bleeding (OR 5.71, 95% CI 2.17–15.01, and OR 5.49, 95% CI 1.79–16.85, respectively). In the non-selective NSAID group the risk for upper gastrointestinal bleeding did not reach statistical significance, whereas the risk for lower gastrointestinal bleeding was almost 4-fold and in the CYP2C9 inhibitor group over 3-fold compared to control (OR 3.96, 95% CI 1.13–13.93, and OR 3.12, 95% CI 1.04–9.28, respectively). The incidence of intracranial bleedings was somewhat higher in some of the warfarin interaction groups compared with the control group, but due to the small number of cases (in platelet aggregation inhibitor and coxib groups n=0) we could not demonstrate a statistical significance for this difference (Table III).

Laboratory values were available for 85%–95% of all treatment periods depending on the study group and the laboratory variable. In all the interaction groups both the mean and the minimum hemoglobin concentrations were significantly lower compared to the control group (Table IV). This observation was compatible with the means of other red cell indices (MCH and MCV), hematocrit, and erythrocyte count. The mean platelet count was higher in all interaction groups compared to controls; a statistically significant difference was seen in the non-selective NSAID and CYP2C9 inhibitor groups. The mean INR value differed statistically significantly from the control in the CYP2C9 inhibitor group (yet staying in the therapeutic range), but not in the other interaction groups (Table IV).

Table IV.  The means of the laboratory variables.

	Platelet aggregation inhibitor	Antidepressant		NSAID
		SSRI	Non-SSRI	Non-selective	Coxib	CYP2C9 inhibitor	Control
Hb, g/L	128^b	126^b	128 ^a	123^b	124^b	119^b	133
(range)	(89–179)	(77–184)	(91–180)	(70–171)	(85–174)	(60–181)	(78–197)
Min Hb, g/L	115^a	110^b	111^b	107^b	108^b	108^b	121
(range)	(58–175)	(48–184)	(71–172)	(45–167)	(55–173)	(51–181)	(42–197)
HCT	0.38^b	0.38^b	0.38^b	0.37^b	0.37^b	0.36^b	0.39
(range)	(0.26–0.52)	(0.23–0.53)	(0.27–0.53)	(0.20–0.51)	(0.24–0.52)	(0.17–0.56)	(0.24–0.57)
B-Eryt, ×10^12/L	4.14^b	4.11^b	4.15^b	4.05^b	4.07^b	3.91^b	4.32
(range)	(2.61–6.09)	(2.53–5.76)	(2.90–5.65)	(2.37–6.46)	(2.60–5.70)	(2.31–6.68)	(2.49–8.47)
MCH, pg/cell	31	31	31	31	31	31	31
(range)	(21–38)	(24–38)	(25–37)	(23–38)	(25–38)	(22–40)	(18–40)
MCV, fL	92	92	91	91	91	91	91
(range)	(65–136)	(75–111)	(78–108)	(73–109)	(79–110)	(72–117)	(59–118)
B-Throm, ×10^9/L	247^a	244^a	259	250^b	247	259^b	232
(range)	(89–906)	(29–751)	(35–550)	(29–864)	(86–667)	(9–1011)	(9–1223)
INR	2.3	2.4	2.3	2.4	2.3	2.6^b	2.3
(range)	(0.9–4.6)	(0.9–5.5)	(0.9–4.3)	(0.9–5.2)	(0.9–4.6)	(0.9–7.0)	(0.8–7.0)

Adjusted for age, sex, ward, and concomitant exposure to proton pump inhibitor and oral glucocorticoid medications.

SSRI = selective serotonin re-uptake inhibitor; non-SSRI = non-selective serotonin re-uptake inhibitor; NSAID = non-steroidal anti-inflammatory drug; CYP2C9 = cytochrome P450 2C9 isoenzyme; Hb = hemoglobin; HCT = hematocrit; B-Eryt = erythrocyte count; MCH = mean cellular hemoglobin; MCV = mean corpuscular volume; B-Throm = platelet count; INR = international normalized ratio.

^aP<0.05 for comparison to the control.

^bP<0.001 for comparison to the control.

The proportion (%) of treatment periods during which the mean hemoglobin concentrations were below the lower limit of the target range (117 g/L for women, 134 g/L for men) was the largest in the CYP2C9 inhibitor group (57.8%) and the lowest in the control group (27.9%) (Table V). The risk (OR) for hemoglobin concentration being below the target range was 2.85 (95% CI 2.44–3.33) in the CYP2C9 inhibitor group. In approximately 70% of treatment periods in the interaction groups the minimum hemoglobin value was under the lower limit of the target range, whereas the same concerned less than half of the treatment periods in the control group. The platelet count was higher than the upper limit of the target range (360×10⁹/L) in 6.6%–13.5% of interaction and 4.5% of the warfarin-only treatment periods. The mean INR value stayed in the therapeutic range in over 90% of all warfarin treatment periods. The only exception was the CYP2C9 inhibitor group, in which 15.7% of the treatment periods showed mean INR values above the therapeutic range (>3.0) yielding nearly a 3-fold risk for too intensive anticoagulation.

Table V.  The proportions of the treatment periods with Hb and Min Hb below, and B-Throm and INR above the target range, and the risk for these extra-target range values.

	Hb			Min Hb			B-Throm			INR
	% of total	OR	(95% CI)	% of total	OR	(95% CI)	% of total	OR	(95% CI)	% of total	OR	(95% CI)
Platelet aggregation inhibitor	39.3	1.50^a	(1.16–1.92)	55.8	1.37^a	(1.06–1.76)	6.8	1.75^a	(1.11–2.74)	7.1	1.14	(0.75–1.72)
Antidepressant
SSRI	43.8	1.71^b	(1.43–2.03)	68.4	2.10^b	(1.74–2.52)	6.6	1.29	(0.93–1.80)	7.1	1.09	(0.80–1.48)
Non-SSRI	36.5	1.11	(0.80–1.54)	70.0	2.11^b	(1.47–3.03)	12.8	2.37^b	(1.48–3.80)	5.9	0.90	(0.49–1.65)
NSAID
Non-selective	49.3	1.90^b	(1.60–2.26)	73.4	2.49^b	(2.08–2.98)	8.1	1.58^b	(1.18–2.12)	7.2	1.07	(0.81–1.42)
Coxib	49.6	1.91^b	(1.41–2.58)	71.1	2.05^b	(1.46–2.87)	6.9	1.29	(0.70–2.38)	6.9	1.07	(0.62–1.82)
CYP2C9 inhibitor	57.8	2.85^b	(2.44–3.33)	71.5	2.43^b	(2.06–2.86)	13.5	3.12^b	(2.41–4.05)	15.7	2.85^b	(2.30–3.54)
Control	27.9			46.7			4.5			5.9

Adjusted for age, sex, ward, and concomitant exposure to proton pump inhibitor and oral glucocorticoid medications.

OR = odds ratio; CI = confidence interval; SSRI = selective serotonin re-uptake inhibitor; non-SSRI = non-selective serotonin re-uptake inhibitor; NSAID = non-steroidal anti-inflammatory drug; CYP2C9 = cytochrome P450 2C9 isoenzyme; Hb = hemoglobin; HCT = hematocrit; B-Eryt = erythrocyte count; MCH = mean cellular hemoglobin; MCV = mean corpuscular volume; B-Throm = platelet count; INR = international normalized ratio.

^aP<0.05 for comparison to the control.

^bP<0.001 for comparison to the control.

Discussion

We found that co-administration of warfarin with potentially interacting drugs is common in hospitalized patients. Nearly 50% of all warfarin-treated patients were exposed to potentially interacting co-medication. A recent study by Snaith et al. demonstrates that close to 70% of out-patients receiving warfarin were prescribed at least one interacting medication [21]. Together these findings emphasize the high frequency of potential drug interactions in warfarinized patients. The larger proportion of treatment periods in women in both antidepressant and NSAID groups may be explained by a higher prevalence of depression and rheumatic diseases in women than in men [22], [23].

Our results indicate that the studied co-medications were associated with an increased risk for bleeding in hospitalized patients on warfarin. Significant increase in the bleeding risk was detected in all interaction groups except for the platelet aggregation inhibitor and the non-SSRI groups. The proportion of PPI users was significantly higher in all interaction groups compared to the control group, which suggests increased gastrointestinal toxicity, but may, on the other hand, lead to under-estimation of the true risk for bleeding caused by the studied co-medications. However, the use of oral glucocorticoids was also significantly more common in the interaction groups compared to the controls which, conversely, may lead to over-estimation of the bleeding risk with the studied combinations. Nevertheless, both of these factors were taken into account as potential confounders in the statistical analyses. Due to the low overall incidence of intracranial bleedings (20 cases in the interaction groups and 8 cases in the control group) no statistically significant increase in the risk for intracranial bleeding associated with warfarin drug interactions could be demonstrated in the present material.

The greatest bleeding risk was associated with co-administration of warfarin with CYP2C9 inhibitors, the risk being 3.6-fold compared to patients receiving warfarin without interacting medication. The risk for upper gastrointestinal bleeding was over 5-fold in the CYP2C9 inhibitor group. This result indicates that although INR monitoring is extensively used in a hospital setting, the increased bleeding risk associated with these interactions may be difficult to control by INR monitoring only. In many cases these co-administrations could have been prevented by, for example, an alternative choice of antimicrobial or antifungal medication.

Concomitant administration of non-selective NSAIDs or coxibs was associated with the overall bleeding risk of about a similar magnitude, 2.6–3.1-fold. Further analysis showed no difference in the risk for all or upper gastrointestinal bleeding between these two groups, confirming the results previously published by Battistella et al. [8]. This result further substantiates that coxibs do not provide increased gastrointestinal safety versus classical NSAIDs in patients taking warfarin.

The use of SSRI antidepressants, but not non-SSRIs, was associated with increased risk for bleeding, especially gastrointestinal bleeding, in warfarinized patients. This is one of the major findings in our study, and it supports the serotonin hypothesis according to which the increased risk for bleeding in the users of antidepressants is strongly associated with serotonin re-uptake inhibition [24], which lowers the platelet serotonin concentration and decreases platelet function [25]. Mirtazapine, the most common non-SSRI (91% of treatment periods) in this study, is not an inhibitor of serotonin transporter, and thus its effect on hemostasis should be lower than that of other antidepressants [26]. Consequently, non-SSRIs might, according to the present results, be a safer alternative to SSRIs with regard to bleeding in patients taking warfarin. However, the number of bleeding diagnoses in the non-SSRI group was low, and further studies are needed to evidence the safety of concomitant non-SSRI therapy in warfarinized patients.

The increased risk for bleeding associated with exposure to warfarin drug interactions shown by the diagnosis data was supported by the laboratory data. Compared with controls, hemoglobin concentrations were significantly lower in all interaction groups, the difference being the largest in the CYP2C9 inhibitor group. The risk for having Min Hb below the target range was 2–2.5-fold higher in patients taking concomitant antidepressants, NSAIDs, or CYP2C9 inhibitors. Although our study detected a difference in the bleeding risk between SSRIs and non-SSRIs, the risk for having Min Hb (but not mean Hb) below the target range was similar in both groups. This may be due to depression leading to, for example, poorer dietary habits independent of which antidepressant medication is used. The mean INR value did not differ from the control in the groups where the interacting drugs affected platelet function. In contrast, in the CYP2C9 inhibitor group the mean INR value was significantly higher compared to the control group, yet staying within the therapeutic range. Still, these patients had nearly a 3-fold risk for over-anticoagulation compared to the control group (Table V). This observation highlights the fact that INR monitoring was not able to detect the increased bleeding risk associated with pharmacodynamic drug interactions altering hemostasis without a change in S-warfarin exposure.

This study was an observational study, and some important limitations should be considered. First, the study population consisted of hospitalized patients, and thus these results cannot be directly generalized to apply to out-patients. Secondly, it is possible, and even likely, that the patients in the interaction groups were generally more ill, as indicated, for example, by significantly higher user rates for PPIs and oral glucocorticoids than those of the control patients, and therefore possessed an increased bleeding risk. Moreover, chronic inflammatory diseases (i.e. the oral glucocorticoid users) are known for lower blood hemoglobin concentration and elevated platelet count [27], [28], which could explain the somewhat higher platelet count and lower mean hemoglobin observed in all interaction groups compared with the control group. Thirdly, although the data used were extracted from a reliable in-house register, in low-dose acetylsalicylic acid–warfarin interaction differentiation between concomitant use and switch may have been compromised. Previously, concurrent acetylsalicylic acid medication with warfarin has been reported to increase bleeding risk 1.43-fold (95% CI 1.00–2.02) by a meta-analysis [29] and 2.75-fold (95% CI 1.44–5.28) by a cohort study [9] compared with warfarin alone. In our study, low-dose acetylsalicylic acid treatment periods comprised 70% of interaction treatment periods in the platelet aggregation inhibitor group. Presumably, a possible switch and its inaccurate recording in the medication registry may explain insignificant differences in the bleeding risks between the platelet aggregation inhibitor and control groups. Such confounding is not likely in other interaction groups.

In conclusion, nearly half of all warfarin-treated in-patients were exposed to potential drug interactions, which were an associated risk for bleedings, especially gastrointestinal bleeding, an inferior hematological status, and a more common use of PPI medication compared to patients on warfarin without interacting medication. Concomitant use of CYP2C9 inhibitors was associated with the highest increase of bleeding risk, and, especially, these risks could have been avoided by an alternative choice of medication. In warfarinized patients, coxibs were associated with a similar risk for bleeding as non-selective NSAIDs, but non-SSRI antidepressants had significantly lower risk for bleedings compared with SSRIs. Co-medication in patients on warfarin should be carefully evaluated to decrease the risk for bleeding.

Acknowledgements

M.H. reports receiving grant support from Clinical Drug Research Graduate School, Helsinki, Finland. T.T. reports receiving grant support from The Clinical Drug Research Graduate School, Helsinki, Finland, The Finnish Cultural Foundation, The Regional Fund on Varsinais-Suomi, and the Research Foundation of Orion Corporation. T.T. and K.L. report receiving Turku University Hospital Grant (grant EVO). The aforementioned funding sources have had no involvement in the study design, in data collection, in analysis or interpretation of data, in writing the manuscript, or in the decision to submit the paper for publication. Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.