Description: The health and cost consequences of individuals receiving a false diagnosis from the index test should be clearly described, including when and how the false result can be identified and corrected.

Rationale: While DTA studies present a direct comparison between the index and the reference tests, economic evaluations tend to model the real-world clinical pathway in which generally only one type of test (index or reference) is received by an individual at the time. This means that in order to accurately reflect the disease progression of individuals post-testing, false test results need to be explicitly modelled. We herein recommend DTA studies to report the health consequences of individuals with a false index test result to facilitate economic modelling. Furthermore, the timing of the potential detection and even correction of these false test results should be described. For instance, individuals with a false negative test result may be detected retroactively via the discovery of unexpected disease progression or exacerbations that are deemed to be attributed to a false diagnosis; the incurred losses of health may be quantified in a DTA study. Individuals with a false positive test may be subject to overtreatment, leading to a decrement in the quality of life.

Example: In a systematic review of economic evaluations on hand-held electrochemical devices for monitoring fractions of exhaled nitric oxide for the diagnosis of asthma (3), the omission of health consequences associated with an incorrect diagnosis was identified to be a major limitation of the published studies. A primary economic evaluation was then performed to overcome this limitation; however, given the paucity of data in the DTA literature, expert opinions were used to determine key model parameters including the number of years required to detect and correct a false diagnosis result, rendering the final modelling findings to be uncertain.

Description: The expected or observed role of an index test within an existing diagnostic pathway and the decision rules if multiple tests are involved should be clearly described.

Rationale: To arrive at a final diagnosis, individuals are sometimes exposed to more than one type of test (11). When an index test is part of a diagnostic pathway with other tests, the DTA study should clarify or propose the role of the index test as the following (12): a replacement (i.e., replacing one or more existing tests due to an increase in diagnostic accuracy), a triage (i.e., preceding the conduct of existing tests for risk stratification so that some individuals may not require the subsequent tests), or an add-on (i.e., to be performed as an additional test in the existing diagnostic pathway). In the presence of multiple tests (11), an ‘OR’ rule may be applicable on some occasions where a positive diagnostic result is reached given any positive individual test result. In other instances, a positive diagnostic result may only be reached if all individual tests are positive (i.e., the ‘AND’ rule). When the results of multiple tests are strongly correlated due to an underlying dependence structure (13), the DTA study may want to report such pairwise correlations and discuss the possibility of omitting one of these tests to streamline the diagnostic pathway.

Example: In 2024, an economic model (4) was published comparing the budget impact of adopting four diagnostic pathways of latent tuberculosis infection involving an interferon-gamma release assay (IGRA) and/or the standard tuberculin skin test (TST), including: i) TST alone; ii) IGRA alone; iii) IGRA following a positive TST result where those tested positive in both tests were deemed to have the disease (i.e., applying the ‘AND’ rule); and iv) TST following a negative IGRA result where individuals receiving a positive result in any test were diagnosed (i.e., applying the ‘OR’ rule). Compared with single-test strategies (i or ii), the ‘AND rule in strategy iii was associated with a lower probability of being treated and lower per-person costs, while the ‘OR’ rule in strategy iv resulted in a higher probability of being treated and higher per-person costs.

Description: For “symptomatic” individuals who receive a negative index test result, DTA studies should explore and report the underlying distribution of other diseases in this patient population.

Rationale: Many symptoms such as cough and headache are associated with more than one possible underlying diseases. As such, an individual in receipt of a correct negative result from the index test (i.e., true negative) may have other undiagnosed health conditions with both cost and quality-of-life implications. In an economic model, these individuals are typically assumed to enter the “healthy (disease-free)” state following a true negative index test result. Until data are made available from DTA studies to illustrate the underlying distribution of diseases in this heterogenous patient population, a comprehensive modelling of the clinical pathway is not feasible.

Example: Asthma, chronic obstructive pulmonary disease (COPD) and congestive heart failure (CHF) are distinct diseases that share many symptoms including the shortness of breath, while the latter two diseases are associated with considerably higher health and cost burdens. In a 2017 economic evaluation on diagnostic tests for asthma (5), the possibility of individuals tested negative for asthma but otherwise had COPD or CHF was entertained in the economic model. However, due to the paucity of objective data on the distribution of COPD/CHF in this patient population, only expert opinions were used to populate the model.

Description: It should be clearly stated if an imperfect reference test (e.g., composite reference standard or other methods) is used to evaluate an index test and how the associated limitations are addressed (if applicable).

Rationale: An imperfect reference test is used to evaluate an index test when a perfect test (i.e., the gold-standard test for determining disease status) is unavailable (14). Due to the inherent limitations of an imperfect reference test, a biased estimation on the accuracy of the index test may arise. For example, despite an index test producing the correct diagnosis for an individual, a DTA study relying on an imperfect reference test may render that diagnosis to be incorrect/false, and thereby introduces errors in the subsequent economic evaluation. To address this limitation, one approach is to assess the index test against a composite reference standard, which classifies individuals into disease positive or negative groups using multiple imperfect tests, though this method has its own limitations related to the uncertainty of the true disease prevalence (15). More sophisticated methods (e.g., latent-class models) are emerging to tackle this issue (14, 15).

Example: Diagnosis of community-acquired pneumonia caused by Streptococcus pneumoniae in adults is often based on a composite reference standard using culture methods (blood culture, sputum, and cultures of other respiratory samples) (10). The BinaxNOW S. pneumoniae (BinaxNOW-SP, the index test) is a urine-based test that can potentially achieve a faster and better diagnosis. In a 2013 review of 27 DTA studies (10), a Bayesian latent-class meta-analysis suggested that the sensitivity of BinaxNOW-SP was higher than the composite reference standard, while both achieved high specificities. An economic evaluation was further conducted using the meta-analyzed test accuracy of the BinaxNOW-SP to improve the precision of the modelling results (6).

Description: DTA studies may want to establish a baseline health-related quality of life (HRQOL) to allow economic evaluations to derive utility values based on the index test result and underlying disease status.

Rationale: Due to the paucity of health utility (HRQOL) data, health economists are confronted with the challenge of assigning a utility value to the population based on the index test result and the underlying disease status (i.e., true/false positive and negative groups; a total of four groups). The common approach is to start with a “baseline” HRQOL value (typically for the true-negative group) using the utility of the general disease-free population (16, 17), and then be able to adjust that utility for the other three groups by incorporating an appropriate decrement. However, this approach has several pitfalls including not being able to account for individuals who tested negative but are otherwise living with other health conditions (see section 3.3) and incorrectly classifying individuals (and thereby assigning the ‘wrong’ utility) due to the use of an imperfect reference test (see section 3.4). Also, the symptom burden for the true-positive and false-negative groups may be different, although both are categorized as having the disease in an economic model and likely will share the same utility value to capture HRQOL. We thereby recommend DTA studies to provide data that can define the baseline HRQOL of the patient population and provide suggestions on the best way to estimate the HRQOL based on the index test result and the underlying disease status.

Example: In the asthma example that we previously presented in section 3.1 (3), individuals receiving a negative index test for asthma (including true and false negatives) were assumed to have the same health utility value as the healthy population, while those who received a false positive result were assumed to experience the same disutility as those with true underlying asthma condition (i.e., true positives). These assumptions may lead to uncertainties in the final modelling results.

Description: Discussion and/or a sensitivity analysis on the limitations associated with a potentially biased disease prevalence estimate should be provided.

Rationale: Disease prevalence is key to determine the cost-effectiveness of an index test compared to the standard of care; however, in many cases the true prevalence of disease cannot be reliably estimated. Specifically, individuals with a negative test result are often not investigated further (e.g., those tested negative in cancer screening), and consequently, the true disease status of these individuals (whether they are a true or false positive in the first place) remains unknown (18), leading to an underestimated disease prevalence. Furthermore, in the absence of a perfect reference standard, the calculation using the accuracy of an index test ([TP + FN]/[TP + FN + FP + TN]) may not reflect the true prevalence of the disease (15). Advanced methods including latent-class modeling (15) are emerging to reduce bias in estimating disease prevalence and may be adopted in the reporting of DTA studies.

Example: An economic evaluation was conducted on noninvasive prenatal testing for trisomy 21 (Down syndrome) at 12-week gestation in 2019 (7). To estimate the true prevalence of trisomy 21 at 12 weeks, the report used the prevalence of trisomy 21 in live birth and the rate of spontaneous pregnancy loss from 12 weeks to term due to the chromosomal anomaly to arrive at an estimation. This estimated prevalence was ultimately used in the economic evaluation.

Description: DTA studies of an AI-enabled index test should report data on model performance over time or comment on the limitations associated with not being able to assess time-dependent accuracy.

Rationale: While most economic evaluations assume the performance of the index test to be constant over the modelled time horizon, there is an emerging class of diagnostic systems powered by AI with improving or degrading performance over time (19, 20). Until adequate information is made available on how often these algorithms are expected to be updated, how these updates will occur, and the time-dependent change of the underlying algorithm (i.e., learning curves), a comprehensive economic evaluation is not feasible (21).

Example: A 2022 evaluation found an AI-assisted computed tomography (CT) scan protocol for lung cancer screening in high-risk adults to be cost-effective over radiologist-interpreted CT scans for the healthcare system over a 20-year time horizon (8). The sensitivity and specificity of the two protocols were obtained from a single DTA study (22) and assumed to remain constant over time. Interestingly, this DTA study also found the underlying deep learning algorithms to be superior to radiologists only during the initial screening, but became on-par when radiologists had access to previous examinations. These results were not incorporated into the economic evaluation and thus may imply the cost-effectiveness results to be overly optimistic favoring the AI-assisted CT scan protocol in a longitudinal screening setting.

Description: DTA studies should present data or comment on the resources and costs of performing the index test.

Rationale: In an economic evaluation, the listing prices of products and services (e.g., technical support and maintenance agreement) obtained from the manufacturers are typically used to estimate the cost of an index test. This method has likely resulted in an underestimation as other cost components including infrastructure depreciation or new purchases, professional fees paid to technicians and other personnel, fees of obtaining the license or specialized software, and other operation costs, are not considered. For example, hospitals that need to establish a new lab (and recruiting personnel) to perform an index test may incur substantial upfront expenditures. We recommend investigators to present this information in a DTA study.

Example: In an economic evaluation of a DNA methylation–based classifier test for central nervous system tumors (9), the capital costs were not included in the reference-case analysis. The inclusion of capital costs (e.g., purchase and equip the entire platform and obtain the license to use the technology) would substantially inflate the cost of testing, and thereby resulting in a less favorable cost-effectiveness profile for the index test. This index test is also powered by a random forest (a type of machine learning algorithm) classification model, which introduces further complexity on how to properly evaluate its cost-effectiveness against the usual care, given the lack of primary DTA data and analytical guidance (see section 3.7).