Based on real-world extensive surveillance activity data, this study revealed that in a cohort of patients with stage I–III breast cancer, more surveillance activities occurred for patients with a predicted lower risk of recurrence compared to higher risk patients. Dependent on the individual, time-dependent and event-specific risk estimations from the INFLUENCE 2.0 nomogram, personalised follow-up schemes potentially reduce the number of clinical visits and associated costs. Our results illustrate the clinical variation and differences in surveillance pathways in the individual care pathways of patients. We observed a slight overutilisation (more than recommended) of breast cancer imaging during follow-up between 2005 and 2020, which was also observed in other studies [12, 26, 27]. Even though the patients in these studies were diagnosed in earlier years and from different hospitals in the Netherlands, the degree of overutilisation is less in this study than previously reported in terms of number of visits (3.9 compared to 1.1 visits in year 1 [26]) and follow-up time (mean 9.0 compared to 4.9 years [27]). Additionally, patients with more than average imaging activities were generally those with a lower risk of recurrence, estimated by the INFLUENCE 2.0 nomogram, as has previously been reported as well [13]. This indicated that without an objective risk estimation with a prediction model, it is difficult for healthcare professionals to assess risk and determine the optimal surveillance strategies for individual patients. Other studies corroborate these findings, stating that the number of observed surveillance visits differed significantly from the guideline [13] and reported the overuse of surveillance visits [26]. One study reported no significant relationship between the way of detection (i.e. routine visits or symptomatic discovery) of the LRR and the grade and stage of the LRR or the risk of DM, further underlining the de-escalation of follow-up [14].

In this cohort, the mean annual risk of recurrence is the highest in the second and third years post-diagnosis, which is consistent with previously reported studies [7, 8, 14, 28, 29]. Cause for the higher risk of recurrence early on in the follow-up period is the fact that more aggressive tumours recur earlier. According to Geurts et al., these types of tumours occur more often in younger (below 40 years), larger tumour size, higher grade, negative hormone status, BCS and > 3 positive lymph nodes [7]. Compared to their cohort, the cohort in this study has more low-grade tumours (60% grade I) and the majority of patients received mastectomy. Therefore, this cohort presumably includes fewer aggressive tumours, which explain why the highest mean annual risk of recurrence occurs in year three instead of the frequently reported second year post-diagnosis.

Moreover, this study has demonstrated that, in line with the guideline, mammography is the mainstay of follow-up care, as nearly every patient (96%) receives at least 1 mammogram during the surveillance period. Additionally, it was shown that a large group of patients receives an annual mammogram over the 5-year period (62%) and thus receives care according to the recommendations of the guideline. However, ultrasound and MRI were also frequently utilised. Yet, the average of 496 days between consecutive ultrasounds indicates that ultrasound may not be part of the annual follow-up strategies. In addition, the average of 1one day between a mammogram and a sequential repeat ultrasound suggests that ultrasound is preferred and performed quickly following an inconclusive mammogram. Although the exact indications and imaging results were not available in the data, one could imagine situations or patients where ultrasound or MRI may have been more suitable than mammography. For instance, for patients with an unfavourable genetic predisposition, screening with MRI is indicated before mammography, and in patients with mastectomy, ultrasound is more appropriate. Despite information regarding the genetic predisposition of patients not being available, patients who received MRI multiple times were generally younger (average 52 years, IQR 45–61). Similarly, the repeat diagnostics shown in Fig. 3 could well be explained by an inconclusive mammogram, after which an additional mammogram or different diagnostic such as ultrasound or MRI would be better suited.

Although the large sample size and level of detail of the data are valuable strengths of this study, the retrospective nature and the unstructured reporting of EHRs may have limited the correct interpretation of surveillance visits and imaging activities. The current completely free-text reporting makes identifying information regarding indication for an imaging activity, result interpretation and patient preferences consistently difficult. Complete and consistent reporting may be achieved through standardised structured reporting [30, 31] and not only eases the reuse of information [32, 33] and decreases of clinical workload [34], but is also directly linked to patient outcomes [35]. Therefore, the reasoning behind interval visits, i.e. whether these were initiated by the patient or the physician and whether the visit was routine or diagnostic was unclear, leading to a possible overestimation of the follow-up.

Furthermore, the INFLUENCE 2.0 nomogram has a defined set of predictors and is based on the NCR and therefore limited to the items collected in the NCR. Additional risk factors may influence the risk of recurrence, such as genetic predisposition or familial history [8]. Expanding the INFLUENCE 2.0 nomogram may increase model flexibility and clinical value. Additionally, other factors such as patient preferences, comorbidities and fear of recurrence could encourage deviating from a follow-up schedule based on the risk predictions by the INFLUENCE 2.0 nomogram [8]. Furthermore, future research into the potential factors contributing to the recurrence rate (e.g. BRCA 1/2, family history and breast density) separately for high-grade and low-grade tumours could be of great value.

Nevertheless, the results in this study outline the use of breast surveillance during follow-up based on real-world data, yet are not directly relatable to what would be a suitable personalised pattern of surveillance. This is mainly because the benefit of annual routine surveillance is still debatable given the many disadvantages regarding false positives, recall rates and increasing the psychological burden on patients, and individual risk predictions are not widely utilised in clinical practice. Despite the guideline recommendation for annual mammography and strong consensus between many different practice guidelines, this routine surveillance is not evidence-based but rather on the assumption that early detection of recurrence reduces breast cancer mortality [36, 37]. Furthermore, evidence of other surveillance intervals or the efficacy of different imaging modalities is lacking, whilst clinical practice is calling for personalised follow-up based on risk and patients’ needs [38]. Withal, this study demonstrates that follow-up is not as straightforward as the guideline describes, actual clinical practice exposes a considerable amount of variation and a large gap between daily practice and risk of recurrence. Personalised surveillance based on objective risk assessment using the INFLUENCE 2.0 nomogram therefore may provide an opportunity to support healthcare professionals in daily decision-making and may increase the efficiency of surveillance strategies.