In situ breast cancer surgeries in Sweden: lumpectomy or mastectomy?—a cost-effectiveness analysis over a 30-Year period using Markov model

This study analysed the cost-effectiveness of current surgical treatments for in situ breast cancer among Swedish breast cancer patients. Sweden is a high-income European country with a robust, predominantly tax-funded, social-based universal health care system [18]. In situ breast cancer, also known as carcinoma in situ of the breast, is coded as D05 in the International Classification of Diseases, Tenth Revision (ICD-10) classification system [19]. This classification system is widely used in national cancer registries, including the Swedish National Cancer Registry (Cancerregistret), which was used for cost estimation in this study [20].

The Swedish National Cancer Registry (Cancerregistret) dataset [20] incorporates information on the ICD-10 code, individual patient surgical and hospitalisation costs. Nevertheless, it lacks specific details on cancer stage and costs associated with surgical complications. To overcome this limitation and to clearly index patients into the model, an additional ICD-10 code (D05), representing carcinoma in situ of the breast, was used. Consequently, the study narrowed its focus to patients with in situ breast cancer.

In situ breast cancer corresponds to the earliest stage of breast cancer, where the malignant cells are still confined to their site of origin. The most prevalent form is ductal carcinoma in situ, characterised by cancerous cells residing in the lining of the breast milk duct, yet not extending beyond the duct into the surrounding breast tissue. Lobular carcinoma in situ, although less common, involves cancerous cells found in the breast’s milk-producing glands (lobules), again confined within the lobules.

In situ breast cancer, often classified as stage 0, usually presents with appropriate resection margins and a ratio of tumour volume to size that enables effective management through various surgical interventions. These interventions include lumpectomy, optionally accompanied by follow-up irradiation or mastectomy. Lymph node removal is not usually needed with lumpectomy. Postsurgical initiation of hormone replacement therapy is an additional treatment consideration, regardless of the surgical approach employed. This study evaluated and compared three available surgical alternatives: mastectomy, lumpectomy without irradiation and lumpectomy with irradiation.

Mastectomy is a type of surgery that involves the removal of the entire breast. Mastectomy may be a treatment option at any stage of breast cancer, including in situ carcinoma depending on the doctor’s recommendation and the patient’s preferences [5, 6, 21]. This surgical method has existed for a long time and is still used in many settings [22].

Lumpectomy, also known as BCS, is a more recent surgical method developed in the early 1980s [23]. This method involves the removal of only a portion of the breast with cancerous cells or abnormal tissue. Follow-up irradiation is typically recommended for patients after a lumpectomy to limit breast cancer recurrence [23]. However, if no clinical value is expected, such as in the case of elderly patients or those with significant health problems, the surgeons may recommend against follow-up irradiation, given the potential for more harm than benefits [24]. The Swedish National Cancer Registry (Cancerregistret) dataset [20] utilised in this study did not stratify the types of follow-up irradiation. As a result, we considered irradiation as a general follow up method to lumpectomy for the purposes of this study.

In adherence to the guidelines provided by the National Board of Health and Welfare, this study was designed to capture the healthcare perspective and only included healthcare costs [25]. Additionally, a broader societal perspective was conducted to account for the indirect costs impacting society at large.

We estimated the difference between the average diagnosis age of in situ breast cancer patients derived from the Swedish National Cancer Registry data in 2020 (Cancerregistret) [20] (~ 58 years). Consequently, a 30-year horizon was chosen to reflect the lifetime approach used in this study.

A six-state Markov model, adapted from Pobiruchin et al. (2016) [27], was implemented in Microsoft Excel (model available upon request). The model [27], previously validated to the German context and built upon a previous model [28], offers several key benefits for our study. Firstly, it accurately reflected the prognosis following initial surgical treatment for our hypothetical cohort of Swedish in situ breast cancer patients, assuming successful treatment in their first surgery. Secondly, it was designed to utilise real-world data on breast cancer, aligning with our approach of calculating parameters using real-world evidence. Lastly, it differentiated between remission and cancer-free states, enabling us to assign unique utility values to these states.

Figure 1 shows a schematic representation of the model structure. The model commences with all patients achieving a “cancer-free” status (State A) following their initial successful surgical treatment: mastectomy, lumpectomy with irradiation or lumpectomy without irradiation. Given the high success rates of early-stage breast cancer surgeries and similar outcomes of both mastectomy and lumpectomy procedures [9], the model did not consider complications from unsuccessful surgeries.

The remaining health states reflect the progression and prognosis of in situ breast cancer after successful initial treatment: “loco-regional recurrence” (State B), “metastasis” (State D), and “death” (State F). Ipsilateral breast cancer events were categorised as local or regional recurrence (State B), and contralateral breast cancer events as metastases (State D) [29]. Patients entering remission after experiencing recurrence or metastasis transitioned to the respective “remission states” (“remission after metastasis” [State C] and “remission after recurrence” [State E]). Importantly, these states are specific to patients who do not revert to their pre-remission disease state. This design eliminates the transitions from State B (or D) back to State A, thereby simplifying the model. Instances of repeated recurrence (second, third, etc.) were accounted for as a continuation in the respective disease state.

Annual transition probabilities (tp) between states were calculated to represent movements between the six-state Markov model. Due to a specific focus on the Swedish context, a targeted review was conducted to gather evidence primarily from the Swedish or Scandinavian context. However, studies from other contexts were appropriately utilized when evidence from the Swedish context was not available. Notably, the three interventions differed primarily in their transition probabilities from cancer-free (State A) to loco-regional recurrence (State B) due to the difference in recurrence rates. Specifically, we assumed that, among the three event rates originating from the cancer-free (State A), only the recurrence rate from cancer-free (State A) to loco-regional recurrence (State B) can be significantly affected by the primary surgery type undertaken. In contrast, very long-term outcomes such as death from breast cancer and the onset of metastasis were considered not directly influenced by the surgery type following after achieving successful surgical intervention.

From this point on, we used the abbreviated names of the states and transition probabilities in the model to reduce repetition in the text. For example, the transition from cancer-free (State A) to loco-regional recurrence (State B) will be denoted tpA2B. Their full meanings can be found in Fig. 1; Table 1.

There was a significant difference in the loco-regional recurrence rate between lumpectomy with and without irradiation [17]. Consistent with other RCTs from the US and the UK [9, 10], the recurrence rate escalated over the years of follow-up, with a pronounced increase in the first 5 to 10 years, gradually diminishing thereafter [15]. Consequently, to account for this time-dependent transition probability, tpA2B, the follow-up time was divided into three periods: 0–5; 5–10 and 10 + years.

The incidence of loco-regional recurrence in lumpectomy with and without irradiation was collected from Wärnberg et al. (2014) [17], a Swedish RCT with a 20-year follow-up period. In the absence of a relevant Swedish study offering reliable data on mastectomy, the incidence of loco-regional recurrence in mastectomy was based on an RCT from the US [9]. To handle contextual differences, we calculated the incidence rate ratio between mastectomy and lumpectomy without irradiation (at 1.27) in the US study and applied this to our model.

The incidence rates retrieved from the literature were converted into annual transition probabilities using the formula p = 1 – exp(-rt) [30], where p is the transition probability, r is the rate, and t is the period of interest. Since the incidence rate had a unit of “per 1,000 person-years”, the value of t would be one year (to also correspond with the cycle length), and the value of r would be the incidence rate divided by 1000. The transition probabilities are listed in Table 2.

The model identified 16 unique transition probabilities (see Table 1). Seven transition probabilities (tpA2B, tpA2D, tpB2C, tpB2F, tpC2D, tpD2E, and tpD2F) out of the 16 were calculated based on cumulative incidence from clinical trials [9, 17, 31‐35] and converted using the following formula: p = 1 – exp(-rt) [30]. Six transition probabilities were calculated by subtracting the other transition probabilities (tpA2A, tpB2B, tpC2C, tpD2D, tpE2E, and tpF2F) from 100% (see Table 1). The remaining three transition probabilities (tpA2F, tpC2F, and tpE2F) were considered similar to the probabilities of dying without breast cancer and were thus assumed to be equal to the age-specific mortality rate of the general Swedish population in 2020, derived from the World Health Organisation (WHO) website (2021) [36].

In addition, we accounted for the probability of dying from natural causes by adding the age-specific mortality rate of the general Swedish population [36] to all states except State F (death) in every year of follow-up.

Costs were estimated from both the healthcare and societal perspectives, in line with the recommendations from the Second Washington panel and the Swedish Dental and Pharmaceutical Benefits Agency (TLVAR) [37, 38]. All costs were valued in 2020 Swedish Krona (SEK), and were only accumulated in State A, State B, and State D, while remission states (State C and State E), and death state (State F) were assumed to incur no cost. The costs calculation were based on an individual-level patient cost dataset from the 2020 Swedish National Cancer Registry (Cancerregistret) [20] and other relevant supporting evidence [12, 13, 39, 40] (see Table 1). The cancer registry data are available upon request.

Costs from the healthcare perspective included the cost of surgery, hospitalisation, and follow-up irradiation (in case of lumpectomy with irradiation). Notably, adjuvant chemotherapy costs were included within the hospitalisation costs.

Baseline costs (before entering State A) accumulated during the first surgical treatment. Costs of surgery and hospitalisation were calculated based on the individual-level patient dataset from National Cancer Registry data in 2020 [20]. The dataset contained more than 5,000 patients undergoing lumpectomy and mastectomy. After restricting the analysis to only in situ cases, 96 (13.24%) have had a mastectomy, and 629 (86.76%) have had a lumpectomy.

Nevertheless, the dataset did not categorise patients according to whether they had lumpectomy with or without irradiation. To get around this, we identified and used the respective Swedish diagnostic-related group cost for lumpectomy without irradiation of 36,439 SEK in 2020 [41] to separate the two lumpectomy alternatives. Consequently, there were 240 (38.16%) lumpectomy without irradiation cases, and 389 (61.84%) cases with irradiation, among the total 629 cases of lumpectomy.

The average costs of mastectomy, lumpectomy without irradiation, and lumpectomy with irradiation were 87,670 SEK (SE: 3421.64), 27,655 SEK (SE: 424.02), and 50,352 SEK (SE: 638.59), respectively (see Table 1). While treatment costs varied across Sweden, these in-country cost variations were not significant. Additionally, a sensitivity analysis was conducted to account for these cost differences (see the section on sensitivity analysis).

At the loco-regional recurrence state (State B), the patients could receive different treatments based on their previous surgery. Recurrence after mastectomy was treated with adjuvant therapy, including radiation therapy and/or chemotherapy according to the recommended standard treatment guidelines. Radiation therapy costs, on average, 5,523 SEK [41]. The cost of hospitalization was assumed to be similar to a lumpectomy with irradiation, averaging 12,157 SEK per episode, based on the Swedish National Cancer Registry data [20]. Consequently, the total cost of a mastectomy alternative in this state was 17,680 (see Table 1).

Total mastectomy was usually recommended for patients with loco-regional recurrence after lumpectomy. Therefore, the total cost for each of the two lumpectomy alternatives at State B was assumed to be the same as that of a mastectomy episode, at 87,670 SEK.

Metastasis state (State D) accounted for the highest cost due to its complication. The cost of metastasis was assumed to be equal for all three surgical alternatives. We considered the most cost-effective treatment option based on a study by Lidgren et al. [40] at 425,174 SEK in 2005. This cost was calculated from a societal perspective, including the formal healthcare cost and informal care from family and friends. After adjusting for informal care costs and inflation, the healthcare cost for State D was 499,343 SEK (see Table 1).

The consumer price index (CPI) is the most widely used measure of inflation, which measures the overall change in consumer prices over time [42]. We applied the inflation calculator for Sweden provided by the WorldData [43] to convert all prices before 2020 to their relative amounts in 2020.

From the societal perspective, we considered costs associated with transportation, informal care, and productivity gains.

Transportation and informal care cost in the metastatic breast cancer state were estimated to be 8,350 SEK annually in 2005 in Sweden [40], and 10,003 SEK in 2020 after adjusting for inflation. We assumed that this cost was similar to the caregiver cost of the more extended hospitalisation options of a mastectomy and lumpectomy with follow-up radiation therapy. For lumpectomy without irradiation, the cost was assumed to be halved at 5,002 SEK (see Table 1).

Productivity gains were estimated based on a study from Norway [12]. When comparing mastectomy and lumpectomy in the first 10 years of follow-up, the productivity increased by 1% annually [12]. Based on this figure and the average salary of Swedish women in 2020 [36], annual productivity gained equalled 4,104 SEK per person (discounted) in the first 10 years for those receiving lumpectomy as their first surgery (see Table 1).

Health outcomes were estimated in terms of quality-adjusted life years (QALYs) as recommended by TLVAR [44]. To the best of our knowledge, no existing study in Sweden has reported QALY weights between lumpectomy and mastectomy. Therefore, we applied a study from the Norwegian context [12] to derive the QALY weights for health State A. Following successful treatment with mastectomy, lumpectomy without irradiation, or with irradiation, the respective QALY weights were 0.84, 0.87, and 0.87 [12] (see Table 1).

The QALY weights for “loco-regional recurrence state” (State B) and “metastasis state” (State D) were taken from a Swedish study [39] and established as 0.78 and 0.69, respectively. The QALY weight for “remission after loco-regional recurrence state” (State C) was assumed to be the average of States A and B. In contrast, the QALY weight for “remission after metastasis” (State E) was the average of States A and D (see Table 1).

The annual discount rate for both cost and QALYs was set at 3% as recommended by the general guidelines for economic evaluations from TLVAR [44].

The average cost-effectiveness ratios were computed for each surgical strategy by dividing the total 30-year cost by the corresponding QALYs for each procedure. To provide a more comprehensive comparison between the strategies, an incremental cost-effectiveness ratio (ICER) was calculated for each pair of competing strategies. The ICER was computed by dividing the costs difference between the two alternatives (e.g. A and B) by their corresponding difference in effects (QALYs in this study).

According to the recommended guidelines [45], all dominated strategies were excluded from the ICER computation. An intervention is considered dominated when it is less effective and more costly compared to an alternative intervention (“strongly dominated”) or the expected additional benefits (health outcomes) are derived at a higher marginal cost than necessary (“extendedly dominated”) [45].

Our study adhered to the guidelines set by the Swedish National Board of Health and Welfare [25], in which ICERs were segregated into four categories. Costs below 100,000.00 SEK per QALY gained were categorised as low, costs between 100,000.00 and 499,999.00 SEK as moderate, costs between 500,000.00 and 1 million SEK as high, and costs exceeding 1 million SEK (equivalent to US $108,652) per QALY gained as very high.

Probabilistic sensitivity analysis (PSA) was performed to handle uncertainties. All inputs were varied according to the statistical distributions obtained from the patient-level dataset and relevant studies referenced (see Table 1). Utilising Excel VBA, the Monte Carlo simulation technique was used to randomly generate values from the joint distributions of costs, utilities, and transition probabilities across 1,000 iterations.

The probabilistic analysis results were presented using a cost-effectiveness plane (CE plane) and a cost-effectiveness acceptability curve (CEAC). The CE plane graphically depicts the outcomes of 1,000 random simulations, representing the uncertainty surrounding the deterministic cost-effectiveness value. The CEAC, on the other hand, represents the likelihood of one alternative being more cost-effective than another, in relation to a range of potential willingness-to-pay (WTP) thresholds.