Two point-of-care test-based approaches for the exclusion of deep vein thrombosis in general practice: a cost-effectiveness analysis

We developed a retrospective patient-level simulation model in Microsoft Excel (version 16.0.14228.20288). It represents the diagnostic pathways of patients with symptoms of DVT from the moment of presentation at the GP until the (correct) diagnosis is established. We applied this health-economic model to analyze the cost-effectiveness of two POCT-based approaches compared to the most common, laboratory-based, situation. Data for the input parameters was obtained from the literature. The outcome measures are the incremental costs, the incremental health effects expressed in Quality-Adjusted Life Years (QALYs), and if these are non-negligible, the incremental cost-effectiveness ratio (ICER) for the two POCT-based approaches compared to the laboratory-based situation.

Of patients presenting at the GP with suspected DVT, 38.4% are male with an average age of 56.9 [8, 10, 19, 20]. In our model we assumed that 79.3% of GPs correctly apply a CDR [11].¹ On the basis that patients arriving after 13:00 h are directly referred for a CUS, we assumed that 70.8% of patients would arrive in time for a laboratory D-dimer assay [21‐24]. The probabilities of a positive or negative D-dimer test or CUS result were obtained from multiple studies [8, 14, 19, 20, 25‐27]. A weighted estimate of sensitivity and specificity was obtained from a series of POCT and laboratory D-dimer assays in order to get a representative performance estimate of these two D-dimer assays [16, 25, 28‐31]. The impact of a false negative test result, in terms of complications (PE and PTS), as well as accompanying costs and health impacts were included in the model, as were complications due to incorrectly treating patients with false positive D-dimer test results (major bleeding). A six-month time-horizon was used in this model; following an initial DVT episode, a three-month treatment window is recommended, so another 3 months later, we expected only few new complications to occur in both treated and untreated patients that could be traced back to the original DVT [7].

Quality adjusted life years were used as effectiveness outcome. The health states incorporated in the model were: healthy, suffering from DVT, suffering from DVT complicated by PTS, PE, major bleeding or recurrence [5, 6, 32‐34]. As a patient can only have a maximum of one quality of life score per year, the maximum per half year, and thus the upper limit of the 95% CI, was set at 0.5.

In accordance with the Dutch costing manual, we considered all costs from a societal perspective. These start from the moment a patient arrives at a GP until (correct) diagnosis up to 6 months of anticoagulant treatment or the occurrence of potential complications. Costs of both office and phone consultations were obtained from the Dutch costing manual [35]. Diagnostic costs from a blood draw, a laboratory D-dimer test and a CUS were derived from the Dutch Health Care Authority [21]. Costs per D-dimer POCT test were determined by considering procurement costs of the device and price per test (including kits and disposables) and were retrieved from sales brochures and retailers. We also included travel and parking costs [21]. The number of production hours lost due to (diagnosing) DVT and potential complications were derived from literature and were both based on age group, gender and the net labor participation [17, 32, 36, 37]. These were then multiplied by costs per production hour lost determined based on the net labor participation classified per gender and age group of Dutch society [17]. To determine treatment costs and costs of potential complications, we derived an average from the literature [21, 38‐41]. The cost estimation was then converted to 2021 Euros based on the Dutch consumer price index [42].

An overview of the model input parameters for each of the three patient strategies used in the model, costs and quality of life score (i.e., utility values) with their corresponding CIs can be found in Tables 2, 3 and 4, respectively. Assumptions for the model are provided as supplementary material.

A probabilistic sensitivity analysis (PSA) was performed by means of a Monte Carlo simulation (10,000 iterations of from 10,000 patients) to demonstrate the effect of joint uncertainty in the input parameters on the outcomes of the model. An incremental cost-effectiveness plane was used to display the uncertainty in model outcomes.

A one-way sensitivity analysis was performed to determine which parameters substantially influence model outcomes. The impact on the output of each strategy was analyzed by applying the lower and upper limit of the corresponding 95% confidence intervals (CIs) of each parameter in the model. Results are presented in tornado diagrams.

A checklist based on the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) is provided as supplementary material. This list comprises the items that should be included when reporting economic evaluations of health interventions and describes where to find these items in this publication.

The main strength of this study is that it was set up to provide an accurate and detailed illustration of diagnostic work-up after a patient with DVT symptoms presents to the GP. In order to achieve this, we incorporated two important real-life input parameters in our model. The first parameter is the estimated production cost loss per health state based on the literature [17, 32, 36, 37]. Net labor participation per complication and associated health-state, sex and age group were used to accurately represent production losses and related costs in real-life DVT patients in the six-month time horizon during and after DVT diagnosis. The second parameter is the arrival time of patients at the GP as this can greatly affect the chosen diagnostic strategy. As these parameters have an impact on costs, they form significant factors.

A further strength is the conservative assumptions made in the model regarding treatment costs, thus possibly underestimating cost savings. For example, treatment and complication costs were based on treatment for 6 months based on available data, even though only 3 months were suggested by the Dutch College of General Practitioners [7]. Moreover, as the sensitivity and specificity of the current POC test devices are lower than their conventional counterparts, more complications were found in the POCT-based approaches. However, D-dimer POCT test devices are rapidly evolving and their diagnostic performance of D-dimer POC tests will most probably improve in the near future; hence, current assumptions about their diagnostic values will probably underestimate the cost-savings of POCT-based strategies. Combining these factors, we expect more cost savings and less health losses using the POCT-based approaches. Furthermore, in our model we assumed that a POC test at the GP practice would lead to a more time-consuming consultation in the fast-POCT strategy. However, if this test could be embedded in the first consultation, this would reduce consultation costs, resulting in additional cost-savings when using the fast-POCT strategy.

Several limitations need to be addressed. First, some input parameters used in the model were obtained from the literature, while others were based on assumptions. One of these assumptions is the time of arrival at the GP of a patient experiencing symptoms that could indicate a DVT. There is no specific data available on how many patients arrive too late for the GP to obtain laboratory D-dimer results before the end of the working day. This is relevant as the follow-up cannot be done on the same day. Because of this, arrival time greatly affects the GP’s strategy so we chose to include this variable in our model. We performed a scenario analysis in which all patients arrived on time vs. another scenario in which they were directly referred to undergo a CUS. This analysis showed no substantial change in model outcomes. Therefore, it is unlikely that this limitation has affected our conclusions.

In other cost-effectiveness studies, the post-DVT mortality risk was included [39, 40]. However, the relatively low incidence of DVT, combined with a low mortality risk and the negligible increase in false negative D-dimer test results due to POCT, would result in a negligible increase in mortality risk. Thereby, including mortality in this already comprehensive and detailed model would not have altered the results in the six-month time-horizon, or when applying a longer time horizon, and would therefore unnecessarily complicate the model.

Patient-friendliness is an important factor in health care. We were unable to include patient-friendliness in our model, as no quantitative measures are currently available for this item. However, waiting time can be reduced, resulting from (i) a POCT assay being performed during the GP consultation in the fast-POCT strategy and (ii) direct referral to the radiologist and internist in the DVT care pathway. These both substantially contribute to patient-friendliness. Furthermore, once the consultation is completed in the two POCT-based approaches, GPs are no longer responsible for follow-up of their patients, so this will also reduce the risk of errors.

We demonstrated that both the DVT care pathway and the fast-POCT strategy will save costs when compared to the regular strategy. If a final decision between these two strategies has to be made, the size and location of the GP practice should be taken into consideration. Larger practices in remote areas may benefit more from the fast-POCT strategy as the majority of (mostly negative-D-dimer) patients will not be referred; this would greatly reduce their travel time. Moreover, in reality, per-test prices generally tend to decrease with higher volumes. For the same reason, smaller practices may benefit more from the DVT care pathway.