The association of vitamin K status with warfarin sensitivity at the onset of treatment


Mary Cushman, MD, MSc, Department of Medicine, University of Vermont, 208 South Park Drive, Suite 2, Colchester, VT 05446, USA. E-mail:


We investigated the association of vitamin K status with warfarin sensitivity among 40 orthopaedic patients beginning perioperative algorithm-dosed warfarin. Baseline vitamin K status was assessed using plasma vitamin K-1 and vitamin K-1 2,3 epoxide concentrations, and a questionnaire-based estimation of usual vitamin K intake. Warfarin sensitivity was assessed as the increase in the International Normalized Ratio (INR) after two doses of 5 mg of warfarin and as the 4-d accumulation of under-γ-carboxylated prothrombin (PIVKA-II), adjusted for warfarin dose requirement. Multivariate models were used to assess vitamin K variables as predictors of warfarin sensitivity. The mean INR increase was 0·53 U and the mean PIVKA-II increase was 771 ng/ml/mg warfarin. Demographic factors were not associated with warfarin response. For each 1 standard deviation (SD) lower value of plasma vitamin K-1, but not the other vitamin K variables, the INR rose 0·24 U (P ≤ 0·01). A higher usual vitamin K intake and plasma vitamin K-1, and lower plasma vitamin K-1 2,3 epoxide, were all associated with a lower PIVKA-II increase over 4 d. Respective differences in PIVKA-II accumulation per SD increase of each variable were −165, −218 and 236 ng/ml/mg warfarin (all P ≤ 0·05). We concluded that dietary and biochemical measures of vitamin K status were associated with early warfarin sensitivity.

There is high inter- and intraindividual variability in the response to oral anticoagulation with anti-vitamin K agents. This often leads to instability of the International Normalized Ratio (INR) and bleeding complications, especially early in the course of therapy. From the clinical perspective, one proposed explanation for this is variation in nutritional vitamin K-1 status (Booth et al, 1997a), however, there are few experimental studies of this topic (Bach et al, 1996; Lubetsky et al, 1999; Kamali et al, 2000). Administration of pharmacological vitamin K to overanticoagulated patients is widely used to reverse the anticoagulant effect of warfarin (Taylor et al, 1999), but the doses used far surpass typical dietary intake.

Measures of vitamin K nutritional status include direct measurement of the vitamin's concentration in plasma (Davidson & Sadowski, 1997) and assessment of dietary intake (Booth & Suttie, 1998). Plasma vitamin K-1 level is largely predicted by recent intake of vitamin K-1 (Booth et al, 1997b). Vitamin K-1 2,3 epoxide is the inactive form of vitamin K that accumulates during warfarin therapy and can also be measured in plasma. One recent study showed an association of dietary vitamin K intake with warfarin sensitivity (Lubetsky et al, 1999) and one small unblinded study reported that a diet with controlled vitamin K content improved management of patients with poorly controlled anticoagulation (Sorano et al, 1993). A recent cross-sectional report of 73 patients with stable anticoagulation showed that a higher INR was positively associated with plasma vitamin K epoxide levels and negatively associated with plasma vitamin K-1 (Kamali et al, 2000).

The INR is the standard measurement of anticoagulant response. Prolongation of the INR occurs as the fully carboxylated forms of vitamin K-dependent factors are depleted and the under-γ-carboxylated forms (the underactive forms) accumulate. These forms can be specifically measured in plasma using research-based tests. The rate of accumulation of under-γ-carboxylated forms is proportional to their synthesis rates, such that factors with a short half-life and high synthesis rate (e.g. Factor VII) influence the INR earlier in therapy, while factors with a long half-life and lower synthesis rate (e.g. prothrombin) influence the INR later.

We investigated the association of vitamin K-1 status with early warfarin sensitivity in patients beginning perioperative warfarin for orthopaedic surgery. Vitamin K status was assessed using both biochemical measurements of plasma vitamin K and estimation of dietary intake. Warfarin sensitivity was measured using the INR and the concentration of under-γ-carboxylated prothrombin (PIVKA-II).

Patients and methods

Patients Between January 1994 and January 1995, consenting adult men and women undergoing total hip replacement or revision at Fletcher Allen Health Care, the teaching hospital for the University of Vermont, were enrolled in the study. Patients with active liver or renal disease, gastrointestinal malabsorption, regular warfarin therapy, contraindication to warfarin, prolonged antibiotic use and regular use of warfarin-interacting medications were excluded. At the preoperative visit, informed consent was obtained using methods approved by the institutional review committee. At that time, baseline INR was measured (see below).

Warfarin dosage and administration Warfarin was begun the evening prior to admission with a dose of 5 mg and compliance was confirmed that night by telephone. Patients were admitted on the morning of surgery to the General Clinical Research Center. The following data were recorded: age, sex, height, weight, smoking history, alcohol use, medication list and medical history. Inpatient warfarin was managed by pharmacists using an algorithm, with doses administered at 18:00 h each day. On the operative day, all patients received 5 mg of warfarin. The target INR was 1·5–2·2 U. Intraoperative blood loss and perioperative blood product administration were recorded.

Laboratory methods On the morning of surgery and on each subsequent day after an overnight fast, blood was drawn with minimal stasis at 07:30 h. Each day the prothrombin time was determined using thromboplastin C (Baxter-Dade, International Sensitivity Index 1·96, normal range 11·6–13·4 s). For storage, blood for vitamin K determination was processed in the dark. Samples were centrifuged at 45 000 g/min at 4°C, divided into aliquots and stored at −70°C. At the end of the study, vitamin K-1 and vitamin K-1 2,3 epoxide levels were determined in plasma from the operative day using reverse-phase high-performance liquid chromatography (Davidson & Sadowski, 1997). PIVKA-II levels on the operative day and post-operative d 4 were determined by immunoassay using a kit (Asserachrom PIVKA-II, American Bioproducts, Parsippany, NJ, USA) (Grosley et al, 1996). Triglyceride concentration was determined by enzymatic methods using Kodak Ektachem clinical chemistry slides.

Diet During the study, patients were given a diet with a known, constant content of vitamin K-1 (Booth et al, 1997a) equal to or surpassing the United States Recommended Daily Allowance (RDA) of 65–80 μg/d (Food and Nutrition Board, 1989). No dietary supplements were allowed. Food not consumed was weighed after each meal and the daily dietary consumption of vitamin K-1 calculated accordingly.

During hospitalization, a food frequency questionnaire was used to assess the usual daily vitamin K intake in the previous 6 months. The questionnaire was developed using direct measurements of the vitamin K-1 content of foods (Booth et al, 1993, 1995). The questionnaire was primarily self-administered, then reviewed by a dietitian for clarification as needed.

Warfarin sensitivity Two calculated measurements were used to assess warfarin sensitivity. First, to assess early sensitivity, INR change was defined as the difference in INR between the preoperative visit and post-operative d 1. All participants were treated with the same dose of warfarin up to post-operative d 1 (two doses of 5 mg), so that this measure of early sensitivity was independent of differences among patients in warfarin dose. Second, as a direct measure of warfarin action, the PIVKA-II change was defined as the difference in PIVKA-II from the operative day to post-operative d 3 divided by the mean daily warfarin dose administered.

Statistical analysis The spss version 9·0 was used for data analysis. Descriptive characteristics of the subjects were calculated. Variability in the data was expressed as the standard deviation (SD). Linear regression was used to assess bivariate associations between measures of vitamin K status and warfarin sensitivity. To determine the combined influence of vitamin K-related variables on warfarin sensitivity, multivariate linear models were fitted with measures of warfarin sensitivity as dependent variables and adjusted for triglyceride levels. For all regression models to predict warfarin sensitivity, associations were presented as standardized regression estimates. Specifically, the difference in warfarin sensitivity for a 1SD change in each vitamin K variable was calculated as the product of the SD and the beta coefficient for that variable. The P-value denoting statistical significance was 0·05.



Characteristics of the 40 patients are shown in Table I. The mean age was 62 ± 13 years and 45% were men. The majority of patients underwent hip replacement for osteoarthritis, with seven patients having either hip revision or replacement for another indication. A minority had chronic medical conditions such as diabetes or coronary disease, with no subjects stating a history of stroke or congestive heart failure.

Table I.  Patient characteristics.
Study variable (Units)nMean (SD) or frequency (%)Range
Age (years)40 62 (13)28–82
Sex (male)40 18 (45) 
Surgery indication40  
Osteoarthritis  33 (83) 
Other   7 (17) 
Weight (kg)40 81·0 (16·3)51·8–137·4
Current cigarette use (yes)40  9 (23) 
Alcohol (drinks/d)40  1·0 (1·0) 0–3
Diabetes (yes)40  5 (13) 
Coronary artery disease (yes)40  6 (15) 
Triglyceride (mmol/l)38  1·81 (0·90) 0·57–3·79
Creatinine (μmol/l)36 79·6 (17·7)61·9–150·3
2-d INR change39  0·53 (0·39) 0–1·90
4-d PIVKA-II change/warfarin dose (ng/ml/mg)39771 (549)22–2377

Measures of vitamin K-1 status and warfarin sensitivity

The usual vitamin K intake, based on the food frequency questionnaire, was 141 ± 87 μg/d, which was higher than both the measured intake over the first 4 d in the hospital (33 ± 24 μg/d) and the RDA of 65–80 μg/d. The mean daily intake of vitamin K in the hospital ranged from 3 μg on the operative day to 128 μg in the 32 patients still in hospital on post-operative d 5. At baseline, after one dose of warfarin, plasma vitamin K-1 2,3 epoxide was present in slightly higher concentrations than vitamin K-1 (2·11 ± 1·76 nmol/l vs. 2·06 ± 1·25 nmol/l). These levels were higher compared with healthy population values (Davidson & Sadowski, 1997).

Regarding the measurements of warfarin sensitivity, the mean INR rise after two doses of warfarin was 0·53 ± 0·39 U (range, 0–1·9 U). PIVKA-II rose from 16 ng/ml to 2681 ng/ml over 4 d (for reference, the range of baseline PIVKA-II values was 2–746 ng/ml). Accounting for the warfarin dose, the mean PIVKA-II rise was 771 ± 549 ng/ml/mg warfarin. The mean INR at 4 d was 1·81 ± 0·37 U (range 1·00–2·70 U). On this day, 28 out of 40 (70%) patients had achieved the target INR (1·5–2·2 U) and 6 out of 40 (15%) were above the target range (maximum INR 2·7 U). THE PIVKA-II and INR were correlated; using linear regression, for each 500 ng/ml higher PIVKA-II at 4 d, the INR on that day was 0·06 U higher (P < 0·001).

Association of vitamin K-1 and other factors with warfarin sensitivity

Table II shows simple linear regression of baseline measurements predicting warfarin sensitivity. There were no associations of surgical factors, age, sex, body-mass index, weight, or measured vitamin K intake in hospital, with warfarin sensitivity by either measure. A higher usual vitamin K intake, as estimated by the food frequency questionnaire, was significantly associated with a slower rise in PIVKA-II, but not with the change in INR. Higher baseline plasma vitamin K-1 was associated with lower warfarin sensitivity, as assessed by the rise in INR, but not the rise in PIVKA-II. While higher baseline plasma vitamin K-1 2,3 epoxide concentration did not reach statistical significance as a predictor of either measure of warfarin sensitivity, its associations with these were in the opposite direction to vitamin K-1 levels and intake. Higher triglyceride concentration was associated with a more rapid anticoagulant response, only as assessed by PIVKA-II. In the analysis of the 4-d INR response, adjusted for the dose of warfarin, results were similar to those of the 2-d change in INR (data not shown).

Table II. Linear regression to predict warfarin sensitivity.
Charateristic2-d INR change4-d PIVKA-II change/warfarin dose (ng/ml/mg)
(definition or SD)SRESESRESE
  • SRE, standardized regression estimate is the regression coefficient multiplied by the standard deviation of the predictor variable (value in parentheses). The value represents the rise in the outcome variable per 1SD rise of the predictor variable. SE denotes standard error of SRE.

  • *

    P ≤ 0·05.

  • **

    P < 0·01.

Age (13 years)0·090·07103 94
Sex (female)−0·060·13224175
Weight (16 kg)−0·080·06−60 87
Usual vitamin K intake (87 μg/d)−0·080·09−208* 83
Vitamin K-1, d 1 (1·25 nmol/l)−0·14**0·06−41 89
Vitamin K-1 2,3 epoxide, d 1 (1·76 nmol/l)0·010·07151 86
Triglycerides (90 mmol/l)0·090·09207* 97

To determine the overall prediction of warfarin sensitivity by vitamin K status, multivariate linear regression models were constructed and the results are shown in Table III. All models were adjusted for triglyceride concentration because of its known close associations with vitamin K-1 and vitamin K-dependent protein concentrations. Of the three vitamin K status variables, only the baseline concentration of vitamin K-1 was significantly associated with a change in INR (P = 0·003). For a difference representing a 1SD increase in baseline vitamin K-1 concentration, the change in INR was 0·24 U less. The estimated usual intake of vitamin K and plasma vitamin K-1 and K-1 2,3 epoxide concentrations were all associated with the PIVKA-II rise (P for all ≤ 0·05). The magnitude of the associations were similar for vitamin K-1 and vitamin K1 2,3 epoxide, but were in the opposite direction. For each 1SD higher value of vitamin K-1, vitamin K-1 2,3 epoxide and the estimated usual vitamin K intake, the respective PIVKA-II increases were −218 ng/ml/mg, 236 ng/ml/mg and −165 ng/ml/mg warfarin. There was no influence of estimated surgical blood loss or perioperative transfusion administration on these findings (data not shown).

Table III.  Multivariate linear regression models predicting warfarin sensitivity.
Characteristic2-d change in INR* (s)4-d PIVKA-II change/warfarin dose (ng/ml/mg)
  • *

    SRE denotes the standardized regression estimate, which is the product of the standard deviation of each predictor variable and the regression coefficient (e.g. the change in the measure of warfarin sensitivity predicted by each 1SD rise in vitamin K status variables). SE denotes standard error of SRE. The P-values represent the significance level for each predictor variable as a continuous term.

Usual vitamin K intake (87 μg/d)0·050·090·46−165810·05
Vitamin K-1 (1·25 nmol/l)−0·240·080·003−218930·03
Vitamin K-epoxide (1·76 nmol/l)0·080·070·56236840·01
Triglyceride (90 mmol/l)0·200·090·02239940·02
Variability predicted by model 0·29   0·38 

In the multivariate models, 29% and 38% of the variability of anticoagulant response measures was explained by the vitamin K measurements and triglyceride concentration. Adjustment for the baseline INR or PIVKA-II had little influence on associations (data not shown).


The main findings of this study were that, at the onset of warfarin therapy in the post-operative setting, measures of vitamin K status were associated with warfarin sensitivity. In general, patients with lower vitamin K status were more sensitive to warfarin than patients with higher vitamin K status. Associations of vitamin K-1 and vitamin K-1 2,3 epoxide with warfarin sensitivity were in the opposite direction, as recently reported (Kamali et al, 2000). Results support a dietary vitamin K−1 warfarin interaction (Booth et al, 1997a) and extend clinical intervention findings concerning the pharmacological use of vitamin K (Taylor et al, 1999). The findings suggest a hypothesis that assessment of vitamin K status might be helpful in the initial management of oral anticoagulation. Currently, although age and sex are thought to be related to warfarin sensitivity (Gurwitz et al, 1992; James et al, 1992), standardized methods of initiation of warfarin only have a 69% success rate in determining the correct maintenance dose (Doecke et al, 1991). Further, a significant proportion of bleeding complications occur in the early weeks of therapy and techniques to reduce these episodes would improve the overall safety of this drug.

Our results confirm and extend a recent report (Lubetsky et al, 1999). In that 8-week study of patients beginning warfarin, two assessments of vitamin K intake during weeks 2 and 8 were made. Compared with low vitamin K consumption, high consumption was associated with lower warfarin sensitivity measured at 8 weeks. Estimated daily vitamin K intake and age of the study group were similar to our population. Our results provide additional information with biochemical assessment of vitamin K, assessment of a vitamin K-dependent protein (PIVKA-II) and reporting of the association of warfarin sensitivity throughout the range of vitamin K intake.

Our experimental findings also agree with a recent cross-sectional study (Kamali et al, 2000), in which an opposite effect of vitamin K-1 2,3 epoxide and vitamin K-1 on anticoagulant response was observed. The established normal range for vitamin K-1 2,3 epoxide is 0·08 ± 0·09 nmol/l (Davidson & Sadowski, 1997) and the values we observed were more than 20-fold higher, suggesting a rapid accumulation of epoxide after one warfarin dose. This response (and its associated variability) allowed us to document the direction of association of epoxide accumulation with warfarin sensitivity. While the univariate association of vitamin K-1 2,3 epoxide with PIVKA-II response to warfarin was not statistically significant (Table II), accounting for other vitamin K-related variables, this association was significant and in the direction expected based on on the report by Kamali et al (2000).

Increased warfarin sensitivity was reported in association with two genetic polymorphisms of the cytochrome P450 CYP2C9 (Aithal et al, 1999). We did not assess the impact of this mutation in our population as our sample size was small and DNA was not collected. Future studies should address the interaction of vitamin K intake with these allelic variants.

The usual intake of vitamin K was associated with PIVKA-II accumulation but not INR response. This discrepancy may relate to imprecision of the dietary questionnaire and insufficient power. However, previous studies support a closer association of dietary vitamin K intake with PIVKA-II compared with the INR (Booth & Suttie, 1998). Another consideration relates to differential synthesis rates of vitamin K-dependent clotting factors. At the start of warfarin therapy, because it has a high synthesis rate, factor VIIc is the first procoagulant to be produced in sufficient amounts of the under-γ-carboxy form to prolong the INR, even when activity of other vitamin K-dependent factors is normal. Therefore, the factor VIIc response to warfarin is more probably influenced by recent vitamin K intake (e.g. plasma concentrations) than PIVKA-II. Alternatively, prothrombin has a much lower synthesis rate than factor VIIc, so PIVKA-II accumulation during warfarin might relate more to storage vitamin K-1 (e.g. questionnaire-based assessment of long-term intake).

While it is a research tool and not validated for clinical use, under standardized conditions in the research setting, PIVKA-II might be a more precise biological measure of warfarin sensitivity than the INR because it is a direct measurement of warfarin effect on a single protein with a long half-life. High PIVKA-II sensitivity to warfarin has been suggested by others (Bach et al, 1996). Better precision may underlie the reason that all measures of vitamin K were independent predictors of the PIVKA-II, but not the INR response, and why a larger percentage of the variability of PIVKA-II response was explained by the multivariate models.

The main limitation of this study was the small sample size, which might explain some lack of associations of vitamin K, and variables such as age and sex, with warfarin sensitivity. Measurement of baseline INR up to several days prior to surgery may have influenced our ability to accurately assess changes in INR, thus biasing results to the null hypothesis. The generalization of the findings to non-post-operative populations and those on stable anticoagulation requires additional study.

We conclude that biochemical and dietary assessment of vitamin K status are associated with early warfarin sensitivity. Assessment of vitamin K status, perhaps in combination with genetic testing for the P450 CYP2C9 alleles, might allow the initial dose of warfarin to be determined more appropriately in the future. For most institutions, measurement of vitamin K-1 level is not available, therefore further study is required to assess if a rapidly applied and validated measure of intake can predict the level of plasma vitamin K-1 or anticoagulant response.


Funding for this study was provided by United States Public Health Service T3207594-09 (M.C.), K08 HL03618 (M.C.), GCRC MO1-RR109, and USDA ARS Contract #53–1950–00–1 (S.B. and K.D.). We gratefully acknowledge the contributions of James Howe MD, Steven Incavo MD, the staff of the General Clinical Research Center and the technical work of Frances Bhushan. Discussions with Frits Rosendaal, MD, and his reading of the manuscript, are also appreciated.