A decision model to estimate a risk threshold for venous thromboembolism prophylaxis in hospitalized medical patients

Summary

Background

Venous thromboembolism (VTE) is a common preventable condition in medical inpatients. Thromboprophylaxis is recommended for inpatients who are not at low risk of VTE, but no specific risk threshold for prophylaxis has been defined.

Objective

To determine a threshold for prophylaxis based on risk of VTE.

Patients/Methods

We constructed a decision model with a decision-tree following patients for 3 months after hospitalization, and a lifetime Markov model with 3-month cycles. The model tracked symptomatic deep vein thromboses and pulmonary emboli, bleeding events and heparin-induced thrombocytopenia. Long-term complications included recurrent VTE, post-thrombotic syndrome and pulmonary hypertension. For the base case, we considered medical inpatients aged 66 years, having a life expectancy of 13.5 years, VTE risk of 1.4% and bleeding risk of 2.7%. Patients received enoxaparin 40 mg day¹ for prophylaxis.

Results

Assuming a willingness-to-pay (WTP) threshold of $100 000/quality-adjusted life year (QALY), prophylaxis was indicated for an average medical inpatient with a VTE risk of ≥ 1.0% up to 3 months after hospitalization. For the average patient, prophylaxis was not indicated when the bleeding risk was > 8.1%, the patient’s age was > 73.4 years or the cost of enoxaparin exceeded $60/dose. If VTE risk was < 0.26% or bleeding risk was > 19%, the risks of prophylaxis outweighed benefits. The prophylaxis threshold was relatively insensitive to low-molecular-weight heparin cost and bleeding risk, but very sensitive to patient age and life expectancy.

Conclusions

The decision to offer prophylaxis should be personalized based on patient VTE risk, age and life expectancy. At a WTP of $100 000/QALY, prophylaxis is not warranted for most patients with a 3-month VTE risk below 1.0%.

Introduction

Venous thromboembolism (VTE) is a common preventable cause of morbidity and mortality among hospital inpatients [1,2]. Randomized clinical trials (RCTs) have demonstrated that prophylaxis with heparin can reduce the incidence of pulmonary embolism (PE) and deep vein thrombosis (DVT) in medical patients [3]. Prophylaxis is therefore recommended for hospitalized patients who are not at low risk of VTE [4] and is a Joint Commission core quality measure [5].

Because thromboprophylaxis increases the risk of bleeding, not all patients should receive it. Routine assessment of patients’ risk of VTE and bleeding is recommended to guide prophylaxis decisions [4,5]. A number of risk-prediction instruments can assign a probability of VTE [6–8]. However, although some high-risk groups have been identified [9], there is no established definition of ‘high-risk’ and no recommended prophylaxis threshold. Thus, the decision regarding when to institute prophylaxis is subjective, and the chance of receiving prophylaxis depends more on the hospital or the physician than on the patient’s level of risk [10], A better approach would be to vary the threshold depending on the risk of bleeding and perhaps the patient’s life expectancy.

Weighing these factors against each other is complex. Cost-effectiveness modeling offers one approach to identify a prophylaxis threshold that incorporates bleeding risk and life expectancy. Although a number of prior studies have examined the cost-effectiveness of different VTE prophylaxis strategies [11–13], none has established a risk threshold to identify patients most likely to benefit. Using cost-effectiveness analysis, we aimed to determine a threshold for VTE prophylaxis and to explore the impact of bleeding risk and life expectancy on that threshold.

Methods

We constructed a decision analytic model consisting of two consecutive modules, a decision-tree and a Markov model, to compare prophylaxis with low-molecular-weight heparin (LMWH) versus no prophylaxis from the health system perspective (Figure S1). The decision-tree module followed patients up to 3 months after hospitalization, tracking short-term outcomes, including symptomatic VTE and complications of prophylaxis, but not asymptomatic proximal and distal DVT. Compared with no prophylaxis, patients receiving prophylaxis had a higher risk of bleeding but lower risk of developing symptomatic deep vein thrombosis (DVT) and pulmonary emboli (PE) within 90 days. Patients exposed to LMWH were also subject to heparin-induced thrombocytopenia (HIT), which in turn carried a higher risk of VTE.

Patients who survived index VTE events entered a Markov model with 3-month cycles and a lifelong time horizon. The Markov model included health states representing long-term complications, including recurrent VTE (rVTE), post-thrombotic syndrome (PTS), chronic thromboembolic pulmonary hypertension (CTEPH), disability after intracranial bleed and death. After each cycle, patients moved between health states depending on transition probabilities until everyone died or reached age 100 years. Patients who did not develop VTE during index hospitalization were also followed for life in another 3-month-cycle Markov model.

Patients experiencing VTE during the index hospitalization or within 3 months received anticoagulation for 3 months, whereas those with recurrent VTE (rVTE) received anticoagulation until death or a major bleed. Patients on extended anticoagulation had a higher risk of major bleed but a lower risk of rVTE than those who stopped treatment early because of a major bleed [4].

Outcome measures were direct medical costs in 2015 US dollars and quality-adjusted life years (QALYs), both discounted at 3%/year. We calculated the incremental cost-effectiveness ratio (ICER) as incremental costs divided by incremental QALYs between two strategies. Because the USA has no established cost-effectiveness threshold, we relied on the recommendation of the World Health Organization to use a value between one and three times per capita GDP [14]. We chose a round number near the middle of this range, $100 000/QALY, as our willingness-to-pay (WTP) threshold.

Patient population

The target population was hospitalized medical patients. The average age was 66 years [15,16], with a bleeding risk of 2.7% [17] and life expectancy of 13.5 years [18], which is shorter than that of a typical 66-year-old person (19.2 years) [19]. Although the average VTE risk was 1.4% [15] (details below), the risk for individual patients can vary more than 10-fold [8]. Therefore, we focused on estimating the ICERs for various VTE risks. We then conducted sensitivity analyses on other criteria that could be used to select patients for prophylaxis, including age, life expectancy and risk of bleeding.

Prophylaxis strategy

Prophylaxis consisted of 40 mg of enoxaparin daily during hospitalization. Enoxaparin was used to estimate prophylaxis costs because it is the cheapest and most commonly used LMWH in US hospitals [20]. We chose LMWH over unfractionated heparin (UFH) because LMWH is less likely to cause HIT [21], whereas the bleeding risks and efficacy of both classes are similar [17].

Model inputs and assumptions

Data used to estimate transition probabilities, costs and utilities were derived primarily from US-based studies (Table 1).

Table 1.

Model inputs

Variable	Base case	Range for one-way analysis	PSA distribution	Reference
Effectiveness of VTE prophylaxis (relative risk)
Symptomatic DVT	0.651	0.366–1.154	Normal	[17]
PE	0.576	0.438–0.761	Normal
All bleeding events	1.328	1.078–1.63	Normal
Baseline risk of VTE in patients with no prophylaxis
DVT	0.009	0.008–0.011	Beta	[15]
PE	0.005	0.004–0.006	Beta	[15]
Complications of VTE prophylaxis
All bleeding events	0.027	0.022–0.032	Beta	[17]
Major bleed given all bleeding events	0.125	0.064–0.186	Beta	[17]
Heparin-induced thrombocytopenia	0.008	0–0.017	Beta	[24]
VTE given HIT (relative risk)	15.5	3.9–61.3	Beta	[24]
Complications of VTE treatment
Major bleed	0.016	0.013–0.02	Beta	[27]
Minor bleed	0.068	0.03–0.1	Beta	[28]
Intracranial bleed given major bleed	0.125	0.01–0.24	Beta	[52]
Disabled among survivors	0.198	0.174–0.223	Beta	[53]
Effectiveness of extended vs. 3-month anticoagulation (relative risk)				[4]
Recurrent VTE	0.12	0.05–0.25	Normal
Major bleed	2.63	1.02–6.76	Normal
Recurrent VTE				[25]
Months 1–6	0.049	0.039–0.058	Beta
Months 7–12	0.027	0.025–0.03	Beta
Months 13–60	0.019	0.017–0.021	Beta
Months 61–120	0.010	0.007–0.013	Beta
Months 121+	0	NA	NA
PE given recurrent VTE	0.192	0.105–0.28	Beta	[26]
Post-thrombotic syndrome, all				[26]
Months 1–12	0.042	0.036–0.057	Beta
Months 13–24	0.016	0.01–0.013	Beta
Months 25–60	0.006	0.005–0.007	Beta
Months 61–96	0.001	0.001–0.002	Beta
Months 97+	0	NA	NA
Post-thrombotic syndrome, severe				[26]
Months 1–12	0.006	0.002–0.011	Beta
Months 13–60	0.004	0.003–0.005	Beta
Months 61+	0	NA	NA
CTEPH				[54]
Months 1–24	0.005	0.001–0.008	Beta
Months 25+	0	NA	NA
Mortality
Overall mortality for medical inpatients	0.061	0.0605–0.0611	Beta	[15]
PE during hospitalization	0.095	0.066–0.124	Beta	[55]
Index PE	0.042	0.039–0.052	Beta	[55]
Recurrent PE	0.301	0.123–0.518	Beta	[27]
Recurrent DVT	0.090	0.061–0.116	Beta	[27]
Intracranial bleed	0.183	0.146–0.223	Beta	[56]
Extracranial bleed	0.039	0.027–0.054	Beta	[52]
Patient’s age (years)	66	50–80	NA	[15]
Life expectancy (years)	13.5	2.9–13.5	NA	[18,19]
Proportion of recurrent VTE patients treated as outpatients
Recurrent DVT	0.3	0.2–0.5	Beta	[29,35]
Recurrent PE	0	NA	NA	[57]
Utility
Warfarin use	0.987	0.92–1	Beta	[28]
LWMH use	0.992	0.94–1	Beta	[28]
Disabled from intracranial bleed	0.6	0.02–1.0	Beta	[58]
PTS
Mild/moderate	0.98	0.97–0.99	Beta	[58]
Severe	0.93	0.912–0.948	Beta	[58]
CTEPH	0.88	0.77–1.0	Beta	[59]
QALY loss (days)				[30] and adapted from [29]
DVT	4.5	4.2–4.7	Normal
PE	4.9	4.2–5.4	Normal
Complications of VTE prophylaxis
Extracranial bleed	2.5	0–5	Normal
Intracranial bleed	5.5	0–8	Normal
Minor bleed	0.5	0–1	Normal	Assumption
HIT	8	5.0–10.0	Normal
Complications of VTE treatment
Extracranial bleed	4.7	3.6–5.2	Normal
Intracranial bleed	8.1	5.1–10.3	Normal
Minor bleed	0.5	0–1	Normal	Assumption
Costs ($)
Enoxaparin (per 40-mg dose)	25	9.53–35.68	Gamma	[32]
Nursing and pharmacy costs	49	25–74	Gamma	[33]
Warfarin cost	108	54–163	Gamma
Warfarin	59	NA	NA	[32]
Prothrombin time test	50	NA	NA	[33]
DVT, inpatient	9320	4660–13 981	Gamma	[30]
DVT, outpatient	1375	687–2062	Gamma	[29]
PE	11 383	5692–17 075	Gamma	[30]
Complications of VTE prophylaxis				[30]
Minor bleed	1292	646–1937	Gamma
Extracranial bleed	6458	3229–9686	Gamma
Intracranial bleed	10 332	5166–15 498	Gamma
HIT	20 664	10 332–30 996	Gamma
Complications of VTE treatment				[30]
Minor bleed	1292	646–1937	Gamma
Extracranial bleed	12 241	6121–18 362	Gamma
Intracranial bleed	27 521	13 761–41 282	Gamma
Care after hemorrhagic stroke	4858	2429–7287	Gamma	[37]
Post-thrombotic syndrome				[60]
Mild/moderate, first year	359	180–539	Gamma
Mild/moderate, second year and beyond	146	73–219	Gamma
Severe, first year	1635	817–2452	Gamma
Severe, second year and beyond	718	359–1077	Gamma
CTEPH	15 217	7608–22 825	Gamma	[36]

Note: Values are transition probabilities unless specified otherwise. All clinical probabilities are per Markov cycle of 3 months. CTEPH, chronic thromboembolic pulmonary hypertension; DVT, deep vein thromboembolism; HIT, heparin-induced thrombocytopenia; PE, pulmonary embolism; PSA, probabilistic sensitivity analysis; QALY, quality-adjusted life-year; RR, relative risk.

Efficacy and safety of LMWHs

The benefits and harms of prophylaxis were derived from a meta-analysis of 10 RCTs that compared either subcutaneous low-dose UFH, prophylactic doses of any LMWHs (enoxaparin, nadroparin or dalteparin) or fondaparinux with no heparin among hospitalized medical patients [17]. Prophylaxis significantly reduced the risk of PE and increased the risk of all bleeding events compared with no prophylaxis. Prophylaxis also reduced the risk of symptomatic DVT. Although the odds ratio (OR) was not statistically significant, it was significant in meta-analyses, which included asymptomatic DVT [3,22]. Therefore, we used the point estimate for symptomatic DVT in the base case and its 95% confidence interval (CI) in sensitivity analysis.

Patients exposed to heparin may develop HIT, which increases the risk of VTE [23]. The aforementioned meta-analysis did not distinguish VTE cases that were a result of HIT from others [17]. Because we modeled HIT separately, it was necessary to remove these cases in calculating the efficacy of LMWH or risk double-counting HIT cases. Therefore, we adjusted reported ORs by reducing the number of VTE cases in the intervention group of each RCT, assuming HIT risk in the prophylaxis group to be 0.79% and the relative risk (RR) of developing VTE as a result of HIT to be 15.5 [24]. We then recalculated the ORs and 95% CIs for symptomatic DVT and PE and assigned an increased risk of bleeding for the prophylaxis group [17]. All ORs were converted into RRs for inclusion in the model. Finally, we did not assume any additional mortality benefit from prophylaxis other than that resulting from reduction in PE-related death.

Epidemiological parameters

Baseline risks of symptomatic DVT and PE among non-prophylaxis patients were based on a cohort of general medical patients followed for up to 3 months after hospitalization [15]. Because 70% of included patients received pharmacologic prophylaxis, we adjusted incidence rates by inflating VTE cases among those with prophylaxis, assuming observed rates were reduced in proportion to heparin efficacy [17]. This resulted in an aggregate 3-month VTE risk of 1.4%. Bleeding risks were derived from a meta-analysis [17]. The probability of HIT and risk of VTE with HIT were estimated from a prospective cohort [24].

To estimate the incidence of rVTE for up to 10 years, we used data from a prospective cohort of patients following symptomatic VTE [25]. After 10 years, patients were assumed to no longer be at elevated risk. Incidence rates of PTS were based on an observational study that followed patients for 5 years after index DVT [26].

Patients experiencing DVT are at high risk for all-cause mortality. One meta-analysis reported the case-fatality rate of recurrent DVT to be 9% within 3 months, but did not specify if death was directly related to DVT or other underlying diseases [27]. In the base case, we assumed prophylaxis could reduce all excess mortality, but examined this relationship in scenario analysis. Because patients could die from causes other than VTE or complications of prophylaxis and/or anticoagulation treatment, we estimated patient survival in the first 3 months based on observational data from general medical patients [15]. Between 3 months and 15 years, survival of VTE patients was based on long-term complication data in patients with DVT in a lower limb [18]. For patients with no VTE within 3 months, survival was based on the control group in the same study. Survival beyond 15 years was derived from US life tables [19].

Utilities

We assigned a small utility decrement to prophylaxis and anticoagulation [28]. Disutility because of short-term complications of prophylaxis was based on number of additional hospitalization days, assuming a zero utility for those days [29]. To calculate days lost [29], we assumed that complications would occur between day 0 and day 5 of hospitalization [30], with duration equal to the average length of stay (LOS) of the complication plus 2.5 days (one half the average LOS for patients without complications). This translated into an extra 8 days for HIT, 2.5 days for major extracranial bleeds and 4 days for intracranial bleeds. We assumed that minor bleeds increase LOS by 0.5 days. Events after discharge, including VTE and bleeding from VTE treatment, were assigned disutilities equal to mean LOS for that condition. All utilities were adjusted for age [31].

Costs

We estimated the cost of 5 days of enoxaparin using the average wholesale price in RedBook [32], then added nursing and pharmacy-related costs [33]. To estimate the cost of prophylaxis complications, we multiplied the number of additional hospital days incurred by the average cost per day [30]. Patients with index VTE incurred costs of a full hospitalization for VTE, 3 months of warfarin [34] and INR tests [33]. Because 15% of index VTEs occurred during initial hospitalization, we assumed these patients would stay an extra 2.5 days on average, with attendant costs of hospitalization plus warfarin and INR checks. We assumed all recurrent PE patients were admitted to hospital, whereas 30% of recurrent DVT patients were treated as outpatients [29,35]. The cost of outpatient DVT treatment was based on a previous cost-effectiveness analysis [29].

Costs of PTS were derived from a study of DVT after hip replacement surgery. The cost of CTEPH was based on a claims data analysis of a privately insured population [36]. The annual cost for patients disabled after intracranial hemorrhage was derived from Eckman et al. [37].

Analyses

Threshold analysis

We estimated the ICERs at different 3-month VTE risks to identify the threshold at which the ICER for prophylaxis equaled $100 000/QALY. Next, holding the VTE risk constant at 1.4%, we conducted additional threshold analyses for other parameters, including bleeding risk, patient age and cost of prophylaxis. We also examined values of VTE and bleeding risks at which prophylaxis was harmful, producing fewer QALYs than no prophylaxis. Finally, because enoxaparin is no longer on patent and some hospitals may be able to negotiate prices below those featured in RedBook, we estimated the risk threshold at a cost of enoxaparin equal to one-half of the lowest RedBook price ($4.76).

Deterministic sensitivity analysis

Using average values for medical inpatients, we conducted a one-way sensitivity analysis to examine impacts of all input variables on the ICER. We also conducted a scenario analysis for excess mortality risk of recurrent DVT, and two-way sensitivity analyses to determine the threshold for VTE risk as a function of bleeding risk, age or decreased life expectancy.

Probabilistic sensitivity analysis

We performed 5000 iterations of a Monte Carlo simulation and produced a cost-effectiveness acceptability curve showing the probability of each strategy being cost-effective at different WTP thresholds. We included distributions for any parameter that changed the ICER by more than 5% in one-way sensitivity analysis.

Results

VTE risk threshold

Figure 1 displays the ICER as a function of 3-month VTE risk. The ICER surpasses $100 000/QALY when the VTE risk 3 months after hospitalization falls below 1.0%. At a VTE risk of 1.4%, which represents the average for medical inpatients, the base-case ICER was $57120/QALY. When the cost of enoxaparin was lower, the VTE risk threshold was 0.78%.

Fig. 1.

Incremental cost-effectiveness ratio (ICER) of prophylaxis versus no prophylaxis as a function of risk of venous thromboembolism (VTE). Risk of VTE was considered as the sum of risks of symptomatic deep venous thromboembolism and pulmonary embolism. As the VTE risk increased, the ICER decreased. If the VTE risk was greater than 1.0%, the ICER would be less than $100 000/QALY. [Color figure can be viewed at wileyonlinelibrary.com]

Sensitivity analysis

Threshold analysis

If all other factors were held constant, prophylaxis would not be cost-effective if the bleeding risk exceeded 8.1%, the patient’s age was older than 73.4 years or the cost of LWMH exceeded $60/dose. In addition, prophylaxis was harmful when the VTE risk was less than 0.26% or bleeding risk exceeded 19.2%.

Deterministic sensitivity analysis

Figure 2 shows 10 parameters for which one-way variations changed the ICER by ≥ 10%. The three most influential variables were HIT risk, prophylaxis efficacy against symptomatic DVT, and RR of thrombotic events with HIT. If recurrent DVT carried no excess mortality, the threshold would be 1.16%.

Fig. 2.

One-way sensitivity analysis of parameters most influential on the ICER. Ranges are in parentheses, with the leftmost values leading to the leftmost ICERs. The figure was truncated at $150 000/QALY for display purposes. CTEPH, chronic thromboembolic pulmonary hypertension; DVT, deep vein thromboembolism; HIT, heparin-induced thrombocytopenia; ICER, incremental cost-effectiveness ratio; PE, pulmonary embolism; QALY, quality-adjusted life-year; RR, relative risk. [Color figure can be viewed at wileyonlinelibrary.com]

In a two-way sensitivity analysis, as bleeding risk increased, so did the VTE threshold for initiating prophylaxis (Fig. 3A). Similarly, the VTE risk threshold increased with age (Fig. 3B). Prophylaxis was cost-effective for a 50-year-old patient at a VTE risk of 0.67%, whereas an 80-year-old patient required a risk of 2.0% for prophylaxis to be cost-effective. Similarly, the threshold increased for patients with a short life expectancy (Fig. 3C).

Fig. 3.

Two-way sensitivity analysis. The graph represents the frontier between the two strategies at an ICER of $100 000/QALY. Risk of venous thromboembolism (VTE) was considered as the sum of risks of symptomatic deep venous thromboembolism and pulmonary embolism. Letter 9 denotes base case. ICER, incremental cost-effectiveness ratio; QALY, quality-adjusted life-year. [Color figure can be viewed at wileyonlinelibrary.com]

Probabilistic sensitivity analysis

At $100 000/QALY, prophylaxis had a 70% probability of being cost-effective (Fig. 4).

Fig. 4.

Cost-effectiveness acceptability curve. QALY, quality-adjusted life-year. [Color figure can be viewed at wileyonlinelibrary.com]

Discussion

In this modeling study, we found that VTE prophylaxis, although cost-effective for the average medical inpatient, is not warranted for most patients with a VTE risk of less than 1.0% up to 3 months after hospitalization. In addition, because the majority of prophylaxis benefits occur long after hospitalization, for younger patients with a short life expectancy or older patients (> 73 years) at average risk prophylaxis is not usually cost-effective. Because tools are available to estimate the risk of VTE [6–8,38–40] and bleeding [41] in individual patients, use of VTE prophylaxis could be tailored to individual patients.

The finding that cost-effectiveness of therapy depends on life expectancy appears to be novel. Although it is well documented that VTE prophylaxis does not impact mortality, most physicians incorrectly assume that if prophylaxis prevents PE, which is sometimes fatal, it will also reduce overall mortality. However, randomized trial evidence does not support that conclusion [17]. Instead, the major benefits of prophylaxis are in preventing the long-term complications of VTE, including recurrent VTE, pulmonary hypertension and post-thrombotic syndrome. Because these complications take time to occur and have lasting impact on quality of life, life expectancy is an important determinant of their impact. For patients with short life-expectancy, the opportunity to benefit from VTE prophylaxis is small and prophylaxis is unlikely to be cost-effective for these patients.

Sensitivity analysis offered additional insights. Because DVT is more common than PE, the effectiveness of LMWH against symptomatic DVT is an important determinant of the prophylaxis threshold. The effectiveness in medical patients is unknown, and observational studies have questioned whether prophylaxis reduces DVT [15]. If LMWH does not reduce the incidence of DVT, a much higher threshold of VTE risk would be required (4.1%). We also found that the cost of prophylaxis had little impact on the treatment threshold. Similarly, the threshold was relatively insensitive to bleeding risk because most excess bleeding was minor. An individual would need a bleeding risk almost 10 times greater than average for prophylaxis not to be cost-effective.

Previous analyses, all sponsored by industry, have examined the cost-effectiveness of VTE prophylaxis in medical inpatients [11,12,42–44]. All derived prevalence and efficacy from the MEDENOX study [45], a randomized trial of moderate to high-risk patients, who are not reflective of typical medicine inpatients. Because none of these studies examined a prophylaxis threshold, their findings are of limited use. Establishing a risk threshold allows for individualized treatment based on risk, as advocated by the Joint Commission. Although physicians are supposed to use clinical judgment to weigh risks and benefits for individual patients, in practice this is almost impossible because there are so many probabilities and outcomes to consider that the average human brain would be quickly overwhelmed.

A decision analytic model devoid of costs could be used as well, but produces a much lower threshold (0.3%). We chose cost-effectiveness analysis (CEA) to establish the prophylaxis threshold because in the current environment, it is no longer acceptable to promote medical interventions that have limited benefit while ignoring value. We also chose a high willingness-to-pay value to ensure a conservative risk threshold. At a more traditional $50 000/QALY, the prophylaxis threshold would be 1.5%.

Despite a number of risk calculation tools, ‘low risk’ has traditionally been defined by absence of risk factors, not actual risk. In 1992, the Thromboembolic Risk Factors Consensus Group recommended prophylaxis based on VTE risk assessment. Based on observed rates in young surgical patients, low-risk patients were estimated to have < 1% incidence of proximal DVT and < 0.01% incidence of fatal PE [46]. This practical threshold resulted from the inability to discriminate a lower risk group, not a determination of a reasonable trade-off between the benefits and harms of prophylaxis. Because almost all medical inpatients had some risk factors, near universal prophylaxis was recommended [47]. Subsequently, prophylaxis rates climbed above 80%, leading to concerns of over-treatment and potential harm [15,48,49].

In 2012, the American College of Chest Physicians’ (ACCP) guideline on antithrombotic therapy and prevention of thrombosis reiterated its recommendation for pharmacologic prophylaxis for acutely ill hospitalized medical patients at increased risk of thrombosis [50]. The guideline summarized several risk assessment models (RAMs) and acknowledged their limitations, such as complexity or lack of prospective validation. They recommended using the Padua Prediction Score and defined low-risk patients as those with a score < 4. They referred to a prospective validation study with a VTE incidence of 0.3% in low-risk patients vs. 11% in their high-risk counterparts [7]. Such a bimodal distribution of risk has not been observed in US studies [8] and leaves a wide range as an implied threshold. One US validation study found that those with Padua score ≥ 4 had a 90-day risk of 2.3% vs. 0.8% for those with lower scores [51]. This distribution falls on both sides of our risk threshold, although for younger patients, some patients with scores below 4 might still qualify for prophylaxis, revealing the need for a more nuanced approach. Other RAMs including the Kucher [39], IMPROVE [6] and Intermountain [40] produced similar risk distributions [51]. Based on various models, between 10% and 20% of medical inpatients would qualify for prophylaxis at our 1.0% threshold [6–8], substantially fewer than those receiving it in current practice.

Moving forward, newer models could be tested against our threshold. For example, the Kucher model successfully divides patients into higher and lower risk, but even ‘low risk’ patients have an incidence of > 1% [39], implying that they would benefit from prophylaxis. Such a model is less useful than the Padua, IMPROVE or Inter-mountain models, all of which identify a large group of patients with an average risk below 1% [6,40,51]. In addition, when evaluating new models, the 1% threshold could be used to help identify the appropriate cut-point for defining low risk.

Our study had several limitations. First, we based efficacy on a meta-analysis of RCTs in which participants were generally sicker or had higher risks than most medical inpatients. Others have questioned whether VTE prophylaxis is effective for typical patients, many of whom have short LOS. If prophylaxis is less effective, then treatment will require a higher risk threshold. Second, there were limited data to predict long-term complications, including the change in PTS severity over time or the incidence of CTEPH beyond 2 years. Third, we assumed that the relative increase in bleeding as a result of prophylaxis was independent of baseline bleeding risk. Because patients at high risk of bleeding have been excluded from randomized trials, this assumption may not be correct. As a result, our conclusions about the bleeding risk threshold are less certain than conclusions about VTE risk or age thresholds. As noted in the ACCP guidelines, patients at high risk of bleeding should not receive thromboprophylaxis [50]. Finally, we lacked reliable information on whether recurrent DVT causes excess mortality or is merely associated with it. In sensitivity analysis, varying the excess mortality rate from 9% to 0% raised the treatment threshold from 1.0% to 1.16%.

In conclusion, we found that a reasonable 3-month VTE risk threshold for prescribing prophylaxis would be 1%, based on a WTP of $100 000/QALY, although the threshold can be adjusted for age. At an average VTE risk of 1.4%, prophylaxis with LMWHs was generally cost-effective for hospitalized medical patients. However, because most patients have a risk less than 1%, offering prophylaxis to everyone is an inefficient use of resources. Further, for patients with an estimated VTE risk of < 0.26%, prophylaxis is harmful and should be discouraged. Because benefits of prophylaxis occur over years, prophylaxis also may not be warranted for those with a short life expectancy. Finally, the cost-effectiveness of prophylaxis appears relatively insensitive to bleeding risk; therefore, VTE risk could be the sole criterion to initiate prophylaxis, except when the risk of bleeding is very high. Future risk prediction models may be designed with this threshold in mind. Similarly, evidence-based guidelines would benefit from designating explicit risk thresholds to guide patient-centered decision making.

Addendum

P. Le participated extensively in project design, data collection, statistical analysis and interpretation of the data, and writing of the manuscript; K. Martinez participated in critical writing and revising the manuscript; M. Pappas contributed to interpretation of data and writing of the manuscript; M. Rothberg participated substantially in project design, data analysis and interpretation, and manuscript writing and revision.

Supplementary Material

Supplemental

Fig. S1. Decision model diagram.

Essentials.

• Low risk patients don’t require venous thromboembolism (VTE) prophylaxis; low risk is unquantified.

• We used a Markov model to estimate the risk threshold for VTE prophylaxis in medical inpatients.

• Prophylaxis was cost-effective for an average medical patient with a VTE risk of ≥ 1.0%.

• VTE prophylaxis can be personalized based on patient risk and age/life expectancy.