Effectiveness of pharmacist-participated warfarin therapy management: a systematic review and meta-analysis


Nathorn Chaiyakunapruk, Center of Pharmaceutical Outcomes Research (CPOR), Faculty of Pharmaceutical Sciences, Department of Pharmacy Practice, Naresuan University, Phitsanulok, Thailand 65000.
Tel.: +66 55 96 1826; fax: +66 55 96 3731;
E-mail: nui@u.washington.edu, chaiyakunapr@wisc.edu


Summary. Objective: Although pharmacist-participated warfarin therapy management (PWTM) has been accepted and implemented in various parts of the world, the evidence demonstrating the effects of PWTM compared with usual care on clinical outcomes is lacking. We performed a systematic review and meta-analysis to compare the effects of PWTM with usual care on bleeding and thromboembolic outcomes. Methods: We searched MEDLINE, SCOPUS, EMBASE, IPA, CINAHL, Cochrane CENTRAL, Thai Index Medicus and Thai Medical Index, and reference lists of studies, without language restriction. Databases were searched from their inception to July 2009. The studies using warfarin as an anticoagulant with sufficient data for compilation of 2 × 2 tables were included. Both randomized controlled trials (RCTs) and non-RCTs were considered. Two authors independently reviewed each study, assigned quality scores and extracted data for all outcomes using a standardized form. Pooled effect estimates (risk ratio; RR) were obtained using a random effects model. Result: Of 661 articles identified, 24 studies with 728,377 patients were included. In the random-effects meta-analysis of RCTs, the PWTM group had statistically significant effects on the prevention of total bleeding [RR, 0.51; 95% confidence interval (CI), 0.28–0.94]. However, the effects on major bleeding (RR, 0.64; 95% CI, 0.18–2.36), thromboembolic events (RR, 0.79; 95% CI, 0.33–1.93), all-cause mortality (RR, 0.93; 95% CI, 0.41–2.13) and warfarin-related mortality (RR, 0.65; 95% CI, 0.18–2.42) were not significant. Conclusion: Pharmacist’s participation in the management of warfarin therapy significantly reduces total bleeding, with a non-significant trend towards decreases in other warfarin-related complications.


Since its introduction in 1954, warfarin has become the most commonly used oral anticoagulant worldwide. As a result of its narrow therapeutic index, complicated pharmacokinetic and pharmacodynamic profiles and high frequency of drug interactions, treatment with warfarin is difficult to manage and associated with significant problems [1–3]. Bleeding is the major complication of warfarin therapy [2,4]. On the other hand, however, under-dosing of warfarin can elevate the risk of death from thromboembolic events.

Quality of anticoagulation control as commonly expressed by time spent in the therapeutic international normalized ratio (INR) range is crucial to ensure optimal outcome during warfarin therapy. Recent studies showed that high variability of INR is strongly associated with adverse outcomes [5–8]. Multiple factors have been shown to affect anticoagulation control such as warfarin dosing, warfarin dosage preparation [9], drug or food interaction [10,11], patient compliance [12–14], patient knowledge [14,15] and anticoagulation service settings [16–18].

Recent reviews [16–18] evaluating the effect of service settings on anticoagulation control demonstrated that the anticoagulation monitoring service (AMS), a systematic and coordinated service for anticoagulation management, is associated with somewhat better outcomes than those with usual care. This AMS approach is also supported by the American College of Chest Physicians (ACCP) Evidence-Based Clinical Practice Guidelines with a grade 1B recommendation [8]. Different types of healthcare personnel such as physicians, physician assistants, nurse practitioners and pharmacists may operate the AMS, however, the pharmacist is an integral part in the majority of AMS worldwide [13]. A number of studies found that pharmacist-participated warfarin therapy management (PWTM) led to a significant decrease in warfarin-related hospital admission [19], less frequency of drug interaction [20] and a decrease in length of hospital stay [21]. In addition, a significant improvement in patient compliance [15,22], patient knowledge [15], patient and physician satisfaction [23] and anticoagulation control [24] were also reported. However, there has been a paucity of evidence demonstrating the effects of PWTM on hard clinical outcomes particularly bleedings and thromboembolic events. Therefore, the objective of the present study was to evaluate the effects of PWTM compared with usual care on clinical outcomes by a performing systematic review and meta-analysis of existing studies.


Data sources and search strategy

The following databases were systematically searched: MEDLINE, SCOPUS, EMBASE, IPA, CINAHL, Cochrane CENTRAL, Thai Index Medicus and Thai Medical Index. Databases were searched from their inception to July 2009. For the search strategy, we used the Medical Subject Headings (MeSH) ‘warfarin’ and ‘pharmacists’ and explored the key words ‘warfarin’, ‘anticoagulation’ and ‘pharmacist’ with slight modifications based on the sources (search strategy example was shown in Appendix S1). There was no language and study design restriction. References of initially identified articles were examined to identify additional studies that met the selection criteria. Authors and experts in the field were also contacted.

Outcome measures

The outcomes chosen for this analysis were warfarin-associated bleedings, thromboembolic events (TE) and mortality. Bleeding events were categorized as major and total bleedings.

As there were varying definitions of major bleeding from study to study which could potentially affect our findings, we therefore performed two analyzes on bleeding. The first analysis was based on the definition used by each investigator. We then thoroughly examined the definition of each study and grouped studies based on the their concordance with two standard bleeding definitions including the American College of Chest Physicians (ACCP) [8] and International Society on Thrombosis and Haemostasis (ISTH) [25]. After grouping, we performed the second analysis among studies that met and did not meet the two standard definitions. Total bleeding was defined as a total number of bleeding events (either major or minor). In case more than one bleeding episode occurred in a patient during the study period, only the most significant bleeding episode was counted for total bleeding event. TE was defined as any embolic or thrombotic cerebrovascular accident, deep vein thrombosis, pulmonary embolism or other systemic thromboembolic events. Mortality was defined as all-cause mortality (death from any causes during study period) or warfarin-related mortality (death from complications during warfarin therapy).

Study selection

Abstracts of identified articles were screened to exclude studies that clearly did not meet the eligibility criteria outlined below. A full text version of each article that passed abstract screening was obtained and evaluated. The inclusion criteria were studies that: (i) used warfarin as an anticoagulant, (ii) reported the number (or percentage with total number of patients) of bleedings, thromboembolic events, or mortality, (iii) had a control group (with healthcare professionals other than pharmacist as service providers) and (iv) had a pharmacist that participated in warfarin management as the intervention group. Studies that were not original articles such as comments, letters, reviews, meta-analyses, case reports, surveys or editorials were excluded. Studies from the same population (duplicate studies), studies not reporting effect estimates or with insufficient information to compute effect estimates were also excluded.

Data extraction and quality assessment

Two investigators (S.S. and U.P.) independently reviewed each abstract, completed full-text reviews and extracted information from each study for inclusion in the analysis. Data extracted from each study were study design, number and characteristic of patients, number of complications (bleeding, thromboembolism or mortality), effect estimates and their 95% confidence intervals (CIs) or standard errors (or other necessary information required for estimates computation).

The quality of each study was assessed using a standard score by two investigators (S.S. and U.P.) independently. Downs and Black’s [26] quality assessment method was used to evaluate the quality of all randomized and non-randomized comparative studies included in the meta-analysis. Additionally, the Jadad composite scale [27] (range of score quality: 0–2 = low, 3–5 = high) was used to assess the quality of randomized controlled trials (RCTs). Discrepancies between the two investigators were resolved by discussion and consensus.

Data synthesis and analysis

Meta-analysis was performed using bleeding, thromboembolism and dealth as outcomes. RCTs and non-RCTs (including quasi-experimental, before-after and cohort studies) were analyzed separately given the inherent differences between these types of study designs [28]. To compare the effects of PWTM with usual care on clinical outcomes, DerSimonian and Laird random-effects models [29] were used to pool relative risk (RR) estimates for bleeding, thromboembolic events and mortality. Presence of heterogeneity was assessed using Q-statistic [30]. A P-value < 0.10 was considered as evidence of heterogeneity. Heterogeneity was also presented as I2 which determined the degree of variation across studies that resulted from heterogeneity rather than by chance. I2 can be calculated as I2 = 100% ×(Q-df)/Q (Q, Cochrane’s heterogeneity statistics; d.f., degree of freedom) [30]. A percentage of around 25% (I2 = 25%), 50% (I2 = 50%) and 75% (I2 = 75%) indicates low, medium and high heterogeneity, respectively [30]. In the case where heterogeneity existed, an attempt to explore sources of heterogeneity was made. Publication bias was assessed using Begg’s test with visual inspection of the funnel plot (asymmetrical shape indicates an existence of bias) and Egger’s regression asymmetry test. The P-value < 0.05 in publication bias tests was suggestive of publication bias.

Sensitivity and subgroup analysis

In order to evaluate the robustness of our analysis, a number of sensitivity and subgroup analyzes were performed. Those analyzes included pooling model (fixed-effect vs. random effect), study size (n ≤ 100 vs. n > 100), outlier studies, methodological quality score, type of comparator, patient age, patient population, study setting, follow-up period and pharmacists’ activities.


Search result and included study characteristics

A total of 661 articles were identified Fig. 1. After exclusion of duplicate or irrelevant articles, 120 potentially relevant articles (116 plus four additional articles identified by reviewing references of initially identified articles) were retrieved for more detailed evaluation. Ninety-six articles were excluded. Of these, 78 articles did not meet the inclusion criteria, 17 articles were comments, letters, reviews, meta-analyses or editorials, and one article was a duplicated study. The 24 remaining studies [31–54] consisted of five RCTs [31–35], nine non-RCTs (Quasi-experimental studies) [37,39,41,43,45,46,48,51,53] and 10 cohort studies [36,38,40,42,44,47,49,50,52,54], with a total of 728 377 patients included in the meta-analysis.

Figure 1.

 Flow diagram of study identification, inclusion and exclusion.

The characteristics of included studies are shown online (Table S1). Briefly, among the included studies that reported patients age and gender, the average age was 62.5 years old (range from 46 to 80.5 years) and 52.9% of patients were men. Of the 24 studies, eight studies were from acute settings, 12 studies were from ambulatory care settings and four studies were from both settings. For RCTs, all studies were from ambulatory care settings. Seventeen studies enrolled patients who had been using warfarin prior to enrolment whereas the seven remaining studies enrolled only new warfarin users. Three studies included only surgical patients, while the rest included all patient groups. For types of healthcare facility, 16 studies were conducted in tertiary care hospitals (teaching or general hospital) while the rest were conduced in other settings. Pharmacists performed various activities. The three most common pharmacist activities were warfarin dosage adjustment, medication/drug interaction review and providing patient and/or health care provider education.

Quality assessment

The methodological quality of five RCTs included in the meta-analysis was high as shown by a Jadad scale of 3 (scale range of 0–5) and mean Downs score of 24.2 (range from 21 to 29; maximum possible score, 31) (Table S1). All studies had clearly defined eligibility criteria, reason for patient exclusions and pharmacist activities. In addition, all studies adequately described allocation sequence generation and reported statistical methods including sample-size calculation. However, all of them were open-label trials. For 19 non-RCTs studies, the mean Downs score was 16.7 (range from 11 to 22). As expected, quality of the RCTs was better than the non-RCTs (= 0.001 using Wilcoxon’s rank-sum test of Downs score).


Prior to data analysis, heterogeneity on total bleeding among five RCTs was assessed. Evidence of heterogeneity among these studies was found (I2 = 54%; P = 0.068). The potential sources of heterogeneity originated from the study by Katemateegaroon [31]. After this study was excluded, the I2 was 0.0% (= 0.431) for the remaining four RCTs. Analysis of the four RCTs included revealed that PWTM was significantly associated with a 49% reduction in total bleedings (RR, 0.51; 95% CI, 0.28–0.94) compared with usual care without heterogeneity (Fig. 2, Table 1). In 19 non-RCTs, PWTM was significantly associated with a 29% reduction in total bleedings (RR, 0.71; 95%CI, 0.52–0.96; = 0.028) compared with usual care, with high level of heterogeneity (I2 = 77%; < 0.001). No apparent publication bias was found as determined by funnel plots either for the RCTs or non-RCTs (Egger’s test for bias: = 0.959 and 0.144, Begg’s test for bias: = 1.00 and 0.263, respectively).

Figure 2.

 Effect of pharmacist vs. usual care on total bleeding. The diamond indicates the summary risk ratio and 95% confidence internal (CI). The size of the squares is proportional to the reciprocal of the variance of the studies.

Table 1.   Effect of pharmacist-participated warfarin therapy management compared with usual care on warfarin-related complications*
Outcomes, sourceEvents in PWTM/patients (%)Events in UC/patients (%)Risk ratio (95% CI)P-valueI2 (%)P-value for heterogeneity
  1. PWTM, pharmacist-participated warfarin therapy management; UC, usual care. *Between studies heterogeneity was calculated using the Q-statistic under the assumption of homogeneity in study effects. Number of events and percentages are subject to a rounding error.

Total bleedings
 Randomized controlled trials
  Overall14/367 (4)29/368 (8)0.51 (0.28–0.94)0.0190.00.431
  Wilson et al.[32] 2/112 (2) 1/109 (1)1.95 (0.18–21.16)
  Jackson et al.[33] 9/59 (15)25/68 (36)0.41 (0.21–0.82)
  Chan et al.[34] 1/68 (1) 2/69 (3)0.51 (0.05–5.47)
  Lalonde et al.[35] 2/128 (2) 1/122 (1)1.91 (0.18–20.75)
 Non-randomised controlled trials (n = 19)7318/89583 (8)58226/638449 (9)0.71 (0.52–0.96)0.02876.9< 0.001
Major bleedings
 Randomized controlled trials
  Overall 5/367 (1)10/368 (3)0.64 (0.18–2.36)0.50712.20.332
  Wilson et al.[32] 2/112 (2) 1/109 (1)1.95 (0.18–21.16)
  Jackson et al.[33] 1/59 (2) 7/68 (10)0.61 (0.02–1.30)
  Chan et al.[34] 1/68 (1) 2/69 (3)0.51 (0.05–5.47)
  Lalonde et al.[35] 1/128 (1) 0/122 (0)2.86 (0.12–69.55)
 Non-randomized controlled trials (n = 12)49/4619 (1)91/4595 (2)0.49 (0.26–0.93)0.03046.70.044
Thromboembolic events
 Randomized controlled trials
  Overall 8/367 (2)11/368 (3)0.79 (0.33–1.93)0.6100.00.975
  Wilson et al.[32] 1/112 (1) 2/109 (2)0.49 (0.04–5.29)
  Jackson et al.[33] 5/59 (9) 7/68 (10)0.82 (0.28–2.46)
  Chan et al.[34] 1/68 (1) 1/69 (1)1.01 (0.06–15.90)
  Lalonde et al.[35] 1/128 (1) 1/122 (1)0.95 (0.06–15.07)
 Non-randomized controlled trials (n = 15)44/5335 (1)133/5250 (3)0.37 (0.26–0.53)< 0.0013.70.410
All-cause mortality
 Randomized controlled trials
  Overall10/299 (3)11/299 (4)0.93 (0.41–2.13)0.8670.00.766
  Wilson et al.[32] 5/112 (4) 6/109 (6)0.81 (0.25–2.58)
  Jackson et al.[33] 4/59 (7) 5/68 (7)0.92 (0.26–3.28)
  Lalonde et al.[35] 1/128 (1) 0/122 (0)2.86 (0.12–69.55)
 Non-randomized controlled trials (n = 4)5671/88480 (6)44763/633499 (7)0.85 (0.37–1.98)0.71115.70.313
Warfarin-related mortality
 Randomized controlled trials
  Overall 3/171 (2) 6/177 (3)0.65 (0.18–2.42)0.5240.00.378
  Wilson et al.[32] 0/112 (0) 2/109 (2)0.19 (0.01–4.01)
  Jackson et al.[33] 3/59 (5) 4/68 (6)0.86 (0.20–3.71)
 Non-randomized controlled trials (n = 3)1213/88428 (1)10197/633352 (2)0.77 (0.39–1.53)0.4599.80.330

For major bleeding (four RCTs), the risk ratio for PWTM vs. usual care was 0.64 (95% CI, 0.81–2.36; = 0.507) without heterogeneity (Fig. 3, Table 1). In 11 non-RCTs, PWTM was significantly associated with a 51% reduction in major bleedings (RR, 0.49; 95%CI, 0.26–0.93; = 0.030) compared with usual care, with a medium level of heterogeneity (I2 = 46.7%; = 0.044). Funnel plots did not indicate any apparent publication bias for both RCTs and non-RCTs (Egger’s test for bias: = 0.205 and 0.066, Begg’s test for bias: = 0.174 and 0.484, respectively).

Figure 3.

 Effect of pharmacist vs. usual care on major bleeding. The diamond indicates the summary risk ratio and 95% confidence interval (CI). The size of the squares is proportional to the reciprocal of the variance of the studies.


Out of four RCTs, the risk ratio for PWTM vs. usual care on thromboembolic events was 0.79 (95% CI, 0.33–1.93; = 0.610) without heterogeneity (Fig. 4, Table 1). In 15 non-RCTs, PWTM was significantly associated with a 63% reduction in thromboembolic events (RR, 0.37; 95%CI, 0.26–0.53; < 0.001), without heterogeneity (I2 = 3.7%; = 0.410). There was no apparent systematic bias as assessed by funnel plots among either the RCTs or non-RCTs (Egger’s test for bias: = 0.648 and 0.821, Begg’s test for bias: = 1.00 and 1.00, respectively).

Figure 4.

 Effect of pharmacist vs. usual care on thromboembolic events. The diamond indicates the summary risk ratio and 95% confidence interval (CI). The size of the squares is proportional to the reciprocal of the variance of the studies.


Three RCTs provided sufficient data for mortality analysis. The risk ratio for PWTM vs. usual care on all-cause mortality was 0.93 (95%CI, 0.41–2.13; = 0.867) without heterogeneity (Table 1). In four non-RCTs, PWTM was associated with a 15% reduction in all-cause mortality (RR, 0.85; 95% CI, 0.37–1.98; = 0.711), without heterogeneity. There was no apparent systematic bias as assessed by funnel plots among either the RCTs or non-RCTs (Egger’s test for bias: = 0.061 and 0.705, Begg’s test for bias: = 0.117 and 0.497, respectively). Two RCTs provided sufficient data for warfarin-related mortality analysis. Based on these studies, the risk ratio for PWTM vs. usual care was 0.65 (95%CI, 0.18–2.42; = 0.524) without heterogeneity (Table 1). In three non-RCTs, PWTM was associated with a 23% reduction in warfarin-related mortality (RR, 0.77; 95%CI, 0.39–1.53; = 0.459), without heterogeneity. Funnel plots did not indicate any apparent publication bias for both RCTs and non-RCTs (Begg’s test for bias: = 0.317 and 0.602, respectively).

Sensitivity and subgroup analysis

Sensitivity analysis (Table 2) was performed to investigate the robustness of effect estimates. Changing the pooling models (fixed-effects vs. random-effects) did not significantly modify the pooled RRs for all outcomes. For types of comparator (Table S2), there were studies indicating physicians as a comparator while some studies did not provide sufficient information on the detail of the comparator. We therefore performed a sensitivity analysis comparing PWTM with ‘Physician’ or ‘Unclear’ as a comparator group, separately. The results showed that regardless of comparators, the PWTM group tended to experienced less warfarin-related complications. As small studies were more likely to be published if their results showed positive outcomes [55], we therefore performed a sensitivity analysis excluding small studies (studies with sample size of ≤ 100). The result of our analysis showed that exclusion of small studies did not significantly alter the pooled RRs. We performed sensitivity analysis on bleeding based upon the concordance of bleeding definitions used by each investigator with two standard bleeding definitions (Table S3) as described in the method section. We found that the effect estimates of such analysis still showed a trend toward less bleeding with the PWTM group (Table 2).

Table 2.   Sensitivity and subgroup analyzes*
  1. RCTs, randomised controlled trials; N, number of trials; ACCP, American College of Chest Physicians; ISTH, International Society on Thrombosis and Haemostasis; n/a, not available. *Between studies heterogeneity calculated using the Q-statistic test under the assumption of homogeneity in study effects. The comparators were stated as ‘Usual care’ or ‘Standard care’ but no details are available. All RCTs have a sample size of more than 100 patients. §All RCTs have a Downs score of more than 20. All non-RCTs have a Downs score ≤ 20. **All RCTs indicated that comparator groups were physician. Except “Models”, other parameters used random effects model to estimate the pooled RRs.

 Total bleedingsMajor bleedingsThromboembolismAll-cause mortalityWarfarin-related mortality
NRisk ratio (95% CI)NRisk ratio (95% CI)NRisk ratio (95% CI)NRisk ratio (95% CI)NRisk ratio (95% CI)
  Fixed effects model40.53 (0.30–0.96)40.55 (0.19–1.56)50.74 (0.33–1.70)30.95 (0.42–2.15)20.59 (0.17–2.09)
  Random effects model40.51 (0.28–0.94)40.64 (0.18–2.36)50.75 (0.33–1.72)30.93 (0.41–2.13)20.65 (0.18–2.42)
  Fixed effects model190.91 (0.89–0.94)110.54 (0.39–0.76)150.33 (0.24–0.45)40.95 (0.93–0.98)30.89 (0.84–0.95)
  Random effects model190.71 (0.52–0.96)110.49 (0.26–0.93)150.37 (0.26–0.53)40.85 (0.37–1.98)30.77 (0.39–1.53)
Omitted Katemateegaroon [31] in the analysis of RCTs
 Before omitted50.88 (0.40–1.90)n/an/a50.75 (0.33–1.72)n/an/an/an/a
 After omitted40.51 (0.28–0.94)n/an/a40.79 (0.33–1.93)n/an/an/an/a
Omitted Bond [47] in the analysis of Non-RCTs
 Before omitted190.71 (0.52–0.96)n/an/an/an/a40.85 (0.37–1.98)30.77 (0.39–1.53)
 After omitted180.65 (0.43–0.98)n/an/an/an/a30.57 (0.06–5.45)20.19 (0.02–1.64)
Study omitted in Non-RCTs (n ≤ 100)
 Before omitted190.71 (0.52–0.96)110.49 (0.26–0.93)150.37 (0.26–0.53)40.85 (0.37–1.98)30.77 (0.39–1.53)
 After omitted130.73 (0.52–1.02)90.51 (0.26–1.02)110.34 (0.22–0.52)30.87 (0.22–3.40)20.56 (0.10–3.02)
Methodological quality in Non-RCTs§
 Downs score ≤ 20180.71 (0.52–0.98)110.49 (0.26–0.93)140.41 (0.28–0.59)40.85 (0.37–1.98)30.77 (0.39–1.53)
 Downs score > 2010.65 (0.32–1.33)0n/a10.17 (0.06–0.49)0n/a0n/a
Comparator groups in Non-RCTs**
 Physician150.56 (0.33–0.97)90.50 (0.26–0.96)120.38 (0.24–0.59)30.57 (0.06–5.45)20.19 (0.02–1.64)
 Unclear40.92 (0.90–0.94)20.27 (0.01–11.66)30.34 (0.14–0.82)10.95 (0.93–0.98)10.89 (0.84–0.95)
Major bleeding definition
  Met criteria20.99 (0.18–5.34)20.99 (0.18–5.34)n/an/an/an/an/an/a
  Not met criteria20.57 (0.17–1.92)20.52 (0.03–8.20)n/an/an/an/an/an/a
  Met criteria50.82 (0.39–1.75)40.55 (0.20–1.48)n/an/an/an/an/an/a
  Not met criteria140.86 (0.73–1.01)70.40 (0.15–1.11)n/an/an/an/an/an/a
  Met criteria20.99 (0.18–5.34)20.99 (0.18–5.34)n/an/an/an/an/an/a
  Not met criteria20.57 (0.17–1.92)20.52 (0.03–8.20)n/an/an/an/an/an/a
  Met criteria90.76 (0.44–1.34)80.45 (0.19–1.02)n/an/an/an/an/an/a
  Not met criteria100.92 (0.90–0.94)30.66 (0.25–1.75)n/an/an/an/an/an/a

Subgroup analyzes of RCTs were performed to investigate the associations of PWTM with all outcomes within strata of five different study characteristics (data not shown). These five subgroups were patient age (< 65 vs. ≥ 65 years old), patient population (new vs. current warfarin users), setting (tertiary care hospital vs. others), follow-up period (≤ 3 vs. > 3 months) and pharmacist activities. Overall, the effects of PWTM on total bleeding, major bleeding and thromboembolism were consistent across all subgroups.


To the best of our knowledge, this report is the first meta-analysis to evaluate the effect of PWTM on hard clinical outcomes. Based on our meta-analysis of RCTs, which represent evidence of the highest quality, PWTM is significantly associated with improved hard clinical outcomes especially on the reduction of bleeding complications. For thromboembolic events, the effects were not statistically significant but a positive trend towards a reduction was seen. The positive outcomes associated with PWTM were seen regardless of patient age, types of healthcare facility or patient settings. Overall, our results are consistent with published guidelines [8] and provide a strong foundation to support the implementation of AMS with pharmacists as an integral part of such services.

When analysing the data based on the nature of study designs, the effect of PWTM in non-RCTs was somewhat larger and met statistical significance more often compared with that of RCTs. This may be a result of confounding factors commonly presented in non-RCTs. Nevertheless, the facts that findings from analysis of high-quality RCTs showed significant positive results especially with a large reduction (49%) in total bleeding are reassuring that the PWTM has a genuine positive effect on this outcome. For TE, a significant reduction was only seen in the non-RCTs but not with RCTs. With an event rate of approximately 1–3% annually, a large sample size and long follow-up period is necessary to detect these events. With the pooled number of patients of < 1000 along with a relatively short duration of follow-up in the four RCTs included in our meta-analysis, it is not too surprising that there were no significant differences on this outcome. On the other hand, with a pooled number of patients of > 720 000 patients in the non-RCTs, a significant reduction in TE by PWTM was detected.

The strength of our inferences is based on compliance with stringent criteria for performing a rigorous systematic review. These included the use of a prior protocol designed to address a research question; the method used in the identification of relevant studies; no language restrictions; the rigorous assessment of methodological quality of included studies; the exploration of sources of heterogeneity; and the quantitative summarization of the evidence.

Several limitations of this study should be noted. First, our study populations were from various countries and settings. Differences in the system of care, standard of usual care practice and pharmacist activities may be widely different. However, the effects of PWTM in our analysis of RCTs were consistent across studies, which may indicate high generalizability of PWTM to any circumstances. A key important factor that may potentially affect our findings was the type of comparator. Nevertheless, the result of our sensitivity analysis showed a favorable trend towards PWTM regardless of types of comparators.

Second, as we searched for literature in all relevant major databases plus Thai local databases, studies from other countries listed only in their local databases were not extracted and included in the present study. Nevertheless, after using the funnel plot, Begg’s test and Egger’s test, evidence of publication bias was not observed in any outcomes.

Third, various studies may use slightly different definitions of bleedings. Such differences could lead to a distortion of the overall effects of PWTM on bleedings outcomes. However, from the sensitivity analysis using standard bleeding definitions, we found that the effect estimates of such analysis still showed a trend towards less bleeding with the PWTM group.

Fourth, as there was considerable statistical heterogeneity in most meta-analysis of observational studies, our findings should be interpreted with caution. Nevertheless, the fact that the majority of estimates went in the same direction, this may suggest the absence of clinical heterogeneity among the studies. Heterogeneity is somewhat beneficial [56] in that it let us explore insights about modification of apparent associations by various aspects of exposures and patient populations. Investigation of possible sources of heterogeneity using post hoc subgroup analysis provided no plausible explanation because of the similarity of I2 between before and after subgroup analysis. In addition, it is possible that the I2 heterogeneity test may have been influenced by the outlier study [57] and have excessive power when there are studies with large sample size, as was the case with our meta-analysis. For example, if we exclude the outlier study by Poon et al. [52] and the very large study by Bond et al. [47] (n = 717 396) from the analysis, the I2 of total bleeding outcome is reduced from 76.9% to 48.8%. The relation discovered in the process of exploring heterogeneity, however, may be useful in the planning and execution of subsequent studies.

Fifth, as mentioned above, heterogeneity should be investigated. We found that the study by Katemateegaroon [31] was the potential source of heterogeneity in total bleeding outcome, thus we excluded this study from the final analysis in this outcome as suggested elsewhere [58]. The possible reason for this heterogeneity was attributable to a higher number of patients in the intervention group with co-morbid diseases known to increase risk of bleeding, such as history of gastrointestinal bleeding and renal insufficiency, compared with the usual care group (51 vs. 43 patients, respectively) [59,60]. After this study was excluded, the I2 was 0.0% (= 0.431) for the remaining four RCTs. However, to test whether the study could potentially affect thromboembolism outcomes, we performed additional analysis on thromboembolism by omitting this study. We found that the pooled RR for thromboembolism with or without Katemateegaroon’s study was still identical.

Sixth, the number of RCTs available for analysis on the relationship between PWTM and clinical outcomes are too few to provide conclusive evidence. However, the summary effect estimates are consistent among all outcomes and support the positive effects of PWTM.

Seventh, there was a clear distinction between RCTs and non-RCTs on the duration of follow-up and such differences may affect the ability of these trials to detect the effects of PWTM on clinical outcomes. The median follow-up time for RCTs at 3 months (with a range of 3–6 months) was relatively short whereas the follow-up time for non-RCTs ranged from 6 months to 63 months. With subgroup analysis based on follow-up time, we found that the effects of PWTM on bleeding were largely observed in the trials with short follow-time while these effects were not evident in studies with longer follow-up time. On the contrary, the effects of PWTM on TE were clearly shown in studies with a longer follow-up time (RR, 0.29; 95% CI, 0.14–0.61). However, the effect size was smaller in studies with a short follow-up time (RR, 0.50; 95% CI, 0.32–0.78). Consequently, the design of future trials should have a follow-up time of ≥ 6 months to capture all dimensions of PWTM effects on clinical outcomes.

Our results are somewhat generalizable. We combined studies with different patient populations, different countries, different patient settings, different patient ages, different systems of care and different pharmacist activities. Thus, our results may be applied to most hospital settings worldwide. In addition, we recommend pharmacists’ activities should include providing education to patients and/or other healthcare providers, performing a medication review to prevent/manage drug–drug, drug–disease and drug–diet interactions and providing recommendations on warfarin dosage adjustment. These activities may represent the uniqueness of PWTM based on our findings.

In summary, our findings indicate that the implementation of PWTM may significantly reduce the risk of warfarin-related complications particularly total bleedings by approximately 50% with a non-significant trend towards improvement on other outcomes, compared with usual care. Given the extent of benefits of PWTM, policy makers and guideline developers may consider the inclusion of PWTM as an effective measure in the optimal management of warfarin therapy in most hospital settings worldwide. Further benefits of PWTM, for example cost-effectiveness of using PWTM and the effects of PWTM on percentage of time spend on INR therapeutic range, should be investigated further.


Author contribution to the work: S.S., U.P., N.C. conceived and designed the research; S.S. and U.P. searched and acquired the data, handled the funding and drafted the manuscript; S.S., U.P., N.C. and S.N. performed statistical analysis and interpreted the data; U.P., N.C., S.N. and A.S. supervised, made critical revision of the manuscript for important intellectual content and gave final approval of the version to be published.


We would like to thank the Office of the Higher Education Commission, Ministry of Education, Thailand, for supporting by grant fund under the program Strategic Scholarships for Frontier Research Network for the PhD Program Thai Doctoral degree for this research.

Disclosure of Conflicts of Interest

The authors state that they have no conflicts of interest.