3.1 Foundation and Rationale of System Dynamics Modelling

SD simulation was chosen as the primary analysis method due to its systematic approach to analysing causes and effects relationships within feedback structures and its capacity to model the system at the aggregate level over extended time horizons. Although there are several dynamic simulation methodologies, such as discrete event simulation (DES) and agent-based modelling (ABM), which offer distinct benefits, SD may be better suited for modelling the broad-boundary effects of complex and intricate interventions such as MCED testing. This suitability also arises from its reduced data intensity, capability to take advantage of processing both qualitative and quantitative data, and comparatively lower computational demands [24, 29, 30, 32, 33]. Additionally, SD facilitates the analysis of policy implications, making it an advantageous choice in such a context [30]. Moreover, SD has the capacity to capture system intricacies and characterise structures of complex systems, as well as understand their behaviour over time and address any potential modelling challenges such as patient-level processes, test and treatment combinations, diagnostic efficacy, harms and benefit evaluations, the evolution of clinical practice guidelines and preferences of both patients and clinicians faced by the conventional modelling approaches [21, 23, 30, 34, 35]. Moreover, an essential aspect of SD is non-linearity, which is closely linked to the presence of feedback processes, in other words, an effect is rarely only proportional to the cause, while the cause also can be impacted by the effect directly or indirectly [36].

In the literature, SD modelling has been leveraged for cancer screening research, notably in the Karanfil and Sterman (2020) study [32]. Their model illustrates how decision-makers evaluate the trade-offs between harms and benefits, impacting screening protocols, thresholds and efficiency across different demographics.

Given its capabilities, SD modelling has been selected as the ideal method to represent four distinct populations within cancer streams, including both those eligible and ineligible for SOC screening. This approach meticulously addresses overlaps among these groups, where individuals might qualify for multiple SOC screening modalities. Additionally, this modelling simplifies the management of complex interdependencies and facilitates comprehensive scenario analysis, enabling the exploration of diverse policy outcomes and intervention strategies.

The core elements of SD are stocks, feedback, flows and time delays [36,37,38]. Stocks are defined as accruals or critical accumulations in the system, such as people, hospital beds or healthcare trusts. Flows are also known as rates, which feed in and out of their respective ‘stocks’ and have the same units of stocks per time unit. Figure 1 presents a basic stock and flow structure, illustrating inflow and outflow in an SD model through the effect of births and deaths on population dynamics as an example.

Fig. 1

A system dynamics stock and flow structure with an inflow and outflow

3.2 Model Description

A representation of our conceptual model structure is depicted in Fig. 2. The cancer screening SD model was built using VensimTM software (Ventana Systems Inc., Harvard, TA, USA), employing a time step of one-fourth of a year, meaning the model progresses in 3-month intervals. Within the model, the screening population is kept tracked using stocks and flows, moving through the relevant modelled clinical pathway and followed until they are either diagnosed or undiagnosed with cancer. The stock and flow structure was also designed to further differentiate patients on the basis of cancer stage and whether the diagnoses and staging were because of screening initiated by SOC modalities or MECD testing.

Fig. 2

System dynamics conceptual model structure. CRC, colorectal cancer; MCED, multi-cancer early detection; SOC, standard of care

The SD model mirrors Australia’s established population-based cancer screening pathways for colorectal, breast and cervical cancers. While there is no current national SOC screening for lung cancer yet, the modelled pathways were based on the proposed framework presented in the Cancer Australia 2020 report “Report on the Lung Cancer Screening Enquiry” [39]. The model’s boundary commences with the population (either eligible or ineligible for SOC screening) and terminates with the stage at diagnosis (categorised as stage I, II, III, IV or unknown). The fundamental building blocks of our model are stocks and flows (also known as ‘stock and flow structure’), which represent the flow of patients through the modelled screening and clinical pathways for each cancer type. The detailed screening pathways for each cancer stream were further described in the sub-sections below (3.2.1–3.2.5). The comprehensive stock–flow model structure, encompassing pathways for the four cancers under investigation (colorectal, breast, cervical and lung), and a modelled pathway for individuals ineligible for the SOC screening, is shown in Fig. S1 of the Supplementary Appendix.

3.2.1 Colorectal Cancer

Approximately 15,400 cases are projected for colorectal cancer, making it the fourth most frequently diagnosed cancer in Australia in 2023. In the early 2000s, it held the position of being the most diagnosed cancer in the country [40]. The National Health and Medical Research Council (NHRMC) establishes standards and recommendations for the prevention, early detection and management of colorectal cancer, and the National Bowel Cancer Screening Program (NBCSP) was introduced in 2006 [41]. According to updated NHMRC guidelines in 2017, the recommended screening commences with an iFOBT test every 2 years in asymptomatic individuals, starting from age 50 years to age 74 years [41]. Recently, a change has been made to the age criteria for CRC screening; starting from 1 July 2024, individuals aged 45–49 years will also be eligible to participate in the national screening program [42]. CRC is a considerable health burden in Australia and mortality reduction is crucial, while participation rates in the national screening program remain low [41, 43].

Our simulation model structure for colorectal cancer emulates the current clinical screening pathway, sourced from relevant Australian literature (e.g. AIHW data). Patients presenting with positive results consult a specialist for a subsequent colonoscopy referral. Suspicious findings from the colonoscopy will lead to a histological examination for colorectal cancer. Patients with confirmed malignancies (histologically positive) are further assessed for cancer staging, typically through modalities such as magnetic resonance imaging (MRI), CT or PET scans. Subsequently, patients are categorised across the designated cancer stages. The modelled clinical pathways for colorectal cancer using SOC and MCED testing are presented in Fig. 3A and B, respectively.

Fig. 3

A Clinical pathway for colorectal cancer (SOC screening). B Clinical pathway for colorectal cancer (MCED testing). BC, breast cancer; CC, cervical cancer; LC, lung cancer; MCED, multi-cancer early detection

3.2.2 Breast Cancer

Breast cancer stands as the predominant cancer diagnosis among women in Australia. Projections indicate approximately 20,500 new cases of breast cancer in female patients for the year 2023, constituting roughly 28% of the anticipated female cancer diagnoses. Furthermore, it ranks as the second most frequently diagnosed cancer among individuals aged 20–39 years and 60–79 years while holding the primary position among those aged 40–59 years in Australia [40]. Under the SOC screening programme for breast cancer (BreastScreen Australia Program), eligible women over 40 can have a free mammogram every 2 years, and women 50–74 years are actively invited to undergo mammography [44]. Positive mammographic findings necessitate a referral to a breast cancer specialist for biopsy and histological assessment. Post-diagnosis patients are categorised on the basis of the cancer’s stage. The modelled clinical pathways for breast cancer using SOC screening and MCED testing are presented in Fig. 4A and B, respectively.

Fig. 4

A Clinical pathway for breast cancer (SOC screening). B Clinical pathway for breast cancer (MCED testing). CC, cervical cancer; CRC, colorectal cancer; LC, lung cancer; MCED, multi-cancer early detection

3.2.3 Cervical Cancer

The inception of the National Cervical Cancer Screening Program in 1991 resulted in declines in both cervical cancer incidence and mortality. This can be attributed to the program’s capacity to identify pre-cancerous abnormalities, which, if untreated, could develop into cancer [40]. In this program, women and people with a cervix 25–74 years of age are invited to have a cervical screening test every 5 years [45]. In Australia, cervical cancer screening commences with HPV testing featuring partial genotyping, which can be self-collected or acquired through a general practitioner (GP) visit. Individuals yielding positive HPV outcomes are subsequently referred to a specialist for a colposcopy. If these HPV results remain inconclusive, a retest is advised within a span of 6–12 weeks (requiring a fresh sample). A positive colposcopy result then leads to a histological examination. Upon a cervical cancer diagnosis, patients are stratified and classified according to the cancer’s progression stage. The modelled clinical pathways for cervical cancer using SOC and MCED testing are shown in Fig. 5A and B, respectively.

Fig. 5

A Clinical pathway for cervical cancer (SOC screening). HPV, human papillomavirus; LBC, liquid-based cytology. B Clinical pathway for cervical cancer (MCED testing). BC, breast cancer; CRC, colorectal cancer; LC, lung cancer; MCED, multi-cancer early detection

3.2.4 Lung Cancer

Australia’s lung cancer screening program is scheduled for initiation in July 2025, with a government investment of $263.8 million from 2023 to 2024 to implement a National Lung Cancer Screening Program on the basis of the feasibility assessment conducted by Cancer Australia and recommendation by the Medical Services Advisory Committee (MSAC) [7, 46]. This program is anticipated to avert more than 500 annual fatalities attributed to lung cancer [46]. At present, there is not an established SOC screening programme for lung cancer in the country.

The proposed clinical pathway for lung cancer was delineated on the basis of Australian guidelines presented in the MSAC consideration [7] and after consultation with medical experts. Accordingly, eligible patients will undergo biennial (every 2 years) low-dose CT scans. Those identified with low and moderate risk will be advised to repeat the low-dose CT scan at intervals of 3 months and 6 months, respectively. In contrast, high-risk patients will be immediately referred to a specialist. The consulting specialist will then determine the need for further evaluations, such as tissue diagnosis. For lung cancer, possible diagnostic approaches could involve bronchoscopy or transthoracic needle biopsy (reference). Upon diagnosis, further staging assessments, such as CT, PET–CT or tissue diagnoses, are conducted. Patients are then classified across four stages of cancer (from stages 1 to 4) or categorised as ‘stage unknown’. The modelled clinical pathways for lung cancer using SOC and MCED testing are shown in Fig. 6A and B, respectively.

Fig. 6

A Clinical pathway for lung cancer (SOC screening). B Clinical pathway for lung cancer (MCED testing). BC, breast cancer; CC, cervical cancer; CRC, colorectal cancer; LDCT, low-dose computed tomography; MCED, multi-cancer early detection

3.2.5 Population Ineligible for the SOC Screening

It is assumed that individuals not qualifying for SOC screening, however, may choose to undergo MCED testing. Should a signal indicating one of the four cancers screened by SOC be identified (CRC, breast, cervical or lung), patients will transition into the corresponding clinical pathway. Conversely, if a signal for a cancer type not covered by SOC screening is detected, these individuals are assumed to consult a specialist before exiting the system. Owing to the complexities in modelling cancers not encompassed by SOC, no clinical pathways have been developed for these other cancer types. The modelled clinical pathway for the population ineligible for SOC screening is shown in Fig. 7.

Fig. 7

Clinical pathway for ineligible population (SOC screening). BC, breast cancer; CC, cervical cancer; CRC, colorectal cancer; LC, lung cancer; MCED, multi-cancer early detection

3.2.6 MCED Testing as a Confirmatory Diagnostic Tool

To investigate the clinical and economic impact of utilising MCED testing as a confirmatory diagnostic following standard of care (SOC) screening, a model extension is proposed. This addition will incorporate a subsequent MCED testing phase ‘as an additional stock’ to the above-described SOC-modelled pathways; however, it will be applicable only for patients whose initial SOC screenings (for colorectal, breast, cervical and lung cancers) yield inconclusive results or readings below the threshold for a positive diagnosis, according to current practice. The conceptual framework outlining this modified diagnostic pathway is depicted in Fig. 8.

Fig. 8

Modelled pathway for MCED testing as a confirmatory diagnostic tool. MCED, multi-cancer early detection; SOC, standard of care

3.3 Key Model Assumptions

The model is predicated on several critical assumptions outlined as follows:

Patients are modelled to exclusively undergo either SOC screening or MCED testing. The concurrent utilisation of both MCED and SOC screening for individual patients is not captured within the current SD framework.

For individuals who undergo MCED testing and receive a positive signal for one of the four cancers with established SOC screening pathways, it is assumed they will adhere to the clinical protocol associated with SOC screening for further diagnostic confirmation and staging. Conversely, for those with detected signals of other cancer types, it is assumed they will be directed to specialist care, after which they exit the system. This approach stems from the modelling complexities associated with delineating clinical pathways for an exhaustive list of cancer types within the Australian context.

Cancer stage shift was assumed for individuals who received and were diagnosed through the MCED testing. This assumption was based on the premise that patients identified through the MCED testing approach are anticipated to be diagnosed at an earlier cancer stage compared with those diagnosed via the SOC screening method.

3.4 Study Perspective and Time Horizon

This study will be framed within the context of the Australian healthcare system, considering the costs and benefits (including the increase in diagnostic yield and detection rate) of using MCED tests. It will examine the above-mentioned scenarios at different screening intervals by examining the results of one-time screenings over 2 years. Furthermore, this model can be extended to evaluate long-term benefits over a 5-, 10- and 15-year time horizon, such as improvements in survival and the accrual of quality-adjusted life-years (QALYs), using the stage at which patients are diagnosed and the associated costs.

3.5 Input Parameters and Data Sources3.5.1 Clinical Data

All clinical data relevant to colorectal, cervical and breast cancers, such as incidence and prevalence rates, eligible populations and uptake rates, will be extracted from the National Screening Programs monitoring reports published by the Australian Institute of Health and Welfare (AIHW). Missing data will be complemented by expert opinion (academics and clinicians engaged in the field of cancer early detection). Given that currently in Australia there is no national screening program for lung cancer, all clinical data relevant to lung cancer will be based on the assumptions submitted for the reimbursement of the lung cancer screening program submitted to the MSAC as well as from the Lung Cancer Screening Enquiry report, published by Cancer Australia in 2020 [39]. The enquiry was held to assess global evidence on the benefits and harms of lung cancer screening, target population groups and the design and effective delivery of a national lung cancer screening program in the Australian setting, where lung cancer is the leading cause of cancer death, accounting for nearly 20% of all cancer deaths. Diagnostic accuracy for the MCED tests will be obtained from the literature [47]. Various assumptions regarding the adoption rates of the MCED test by clinicians will be explored through distinct scenario analyses planned to be conducted.

3.5.2 Cost Data

Relevant cost data, including the cost of SOC screening, specialist visits and further assessments, will be extracted from the Australian Medicare Benefits Schedule (MBS) [48]. The average cost of MCED testing will be estimated on the basis of the acquisition costs of various tests, which will be sourced from relevant manufacturers’ online sources, and these costs will then be converted into Australian dollars.

3.5.3 MCED Testing Data

The eligible population for this study was defined as those currently eligible for SOC screening. As MCED testing is not yet in Australia, there are no existing data to inform the clinical pathways for MCED testing. Therefore, the positivity rate, defined as the number of MCED-positive cases among the eligible population, will be estimated by summing the true positives (TP) and false positives (FP). The TP and FP values were calculated on the basis of sensitivity, specificity and prevalence using the formulas provided below (1) and (2) [32].

$$TP = Sensitivity\times Prevalence$$

(1)

$$FP = \left(1-Specificity\right)\times \left(1-Prevalence\right)$$

(2)

3.5.4 Stage Shift Estimation

A critical assumption influencing the projected benefits of MCED is its potential to cause a stage shift in cancer diagnosis, specifically, the shift towards earlier-stage detection compared with the current standard-of-care (SOC) screening methods. The estimation of this stage shift is informed by the interception model developed by Hubbell et al. [15], which provides a probability matrix for each cancer type, indicating the likelihood of earlier stage detection through MCED. To evaluate the anticipated stage shift within the Australian healthcare context, these probabilities will be applied to existing Australian cancer stage distribution data. This approach aims to quantify the potential impact of MCED on the stage at diagnosis among Australian patients, thereby offering insights into its clinical utility and potential for improving cancer outcomes in Australia.

3.6 Model Validation3.6.1 Internal Validity

The outputs of the SD model for various SOC screening pathways, which estimate annual cancer diagnoses, were internally validated using data from the AIHW screening program monitoring reports. The findings from this validation effort are shown in Fig. 9. This validation specifically compared the model’s estimates of diagnosed patients with the actual reported numbers for CRC, breast cancer and cervical cancer after the first screening round within 1 year. Patients diagnosed in the second or subsequent screening rounds were excluded from the validation to simplify the validation process and ensure the model’s accuracy.

Fig. 9

Model validation results over 1 year. *Note: Since there is no existing lung cancer screening program in Australia, data on the number of patients diagnosed with lung cancer through such screenings are unavailable

The primary objective of utilising AIHW data in the validation process was to ensure the model’s accuracy, particularly its ability to replicate results consistent with published data. This step was essential for verifying that the model behaves as expected when simulating the patient journey from screening through to diagnosis. However, we recognise the limitations of this approach, especially the risk of the model merely reproducing input data without fully testing its predictive accuracy.

To address this concern, we plan to incorporate additional validation strategies in any future analyses based on this model. These strategies will include cross-validation with external data sources, where independent datasets not used in the model’s construction will be used for validation. This method will help ensure that the model’s predictions are robust and not just a reflection of the input data. Additionally, sensitivity analyses will be conducted by systematically varying key model parameters and inputs, allowing us to evaluate the impact of different assumptions on the model’s outcomes. This approach will provide a deeper understanding of the model’s reliability and stability across various scenarios.

3.6.2 Expert Involvement and Validation of Clinical Pathways

The development of the SD model was driven by a multidisciplinary team of experts, specifically selected for their roles as clinicians and researchers in cancer and precision oncology. This team included oncologists and early career researchers (J.S. and Y.H.T.), who were chosen on the basis of their experience in precision oncology. They played a pivotal role in refining the model and ensuring its clinical relevance by participating in meetings and round table discussions over two sessions. Their contributions were vital, providing expert insights into current diagnostic practices and the practical aspects of cancer screening and detection.

To ensure the relevance and applicability of the model, the clinical pathways proposed within the model were rigorously checked and validated by the expert team. This validation process involved comparing the modelled pathways against those outlined in the Australian Institute of Health and Welfare (AIHW) screening reports. This step was essential in confirming that the model accurately reflects the real-world clinical pathways used in the Australian healthcare system, thereby enhancing the reliability of the stage shift estimations generated by the model.