Surgical process modeling

Clinical motivation

It is possible to appraise certain facets of a surgical intervention objectively. Examples for such objective aspects are the duration of single work-steps or whole intervention phases or whether or not one approach is more effective than another approach. However, the overall evaluation of surgical practice is very intricate, complex, and highly subjective. There is no method available or described in the appertaining literature that allows for an objective and reliable quantification of surgical practice from a process point of view.

Nevertheless, such an appraisal is more important than ever for various reasons. One of these reasons is the context of the reduction of working hours for medical residents according to the changed working time directives of European law, which might induce an extra pressure on the surgeon to perform faster and to complete more interventions in a single working day. Furthermore, the number of medical – and foremost surgical – students is on the rise. This aggravates the already existing problem with providing enough practical training for each of them.

However, the problem expands to other areas as well. The current situation is such that increasing numbers of novel and innovative technologies become available. The implementation of these technologies into the medical routine with the aim of providing the best possible treatment for patients sometimes competes against ethical or financial concerns, especially when these technologies have not yet been subjected to a multicenter cost-benefit-ratio investigation. With regard to ethical concerns, one example would be a new technology that becomes available that would ameliorate the overall outcome of an intervention but that cannot be financed by reimbursements. In this case, the management of a hospital would have to choose between financial loss and ethical behavior. Effectively, it has to be stated that the overall standard of surgical training faces some serious issues, such as heightened workloads for surgeons, a need for more efficient training structures, increasing complexity of medical technological systems, increasing economic pressure, and a decreasing amount of time-per-patient.

Today, there is an urgent necessity to record, model, and analyze surgical practice by means of computers. This recording technology has to be objective and unbiased and needs to function robustly and unfalteringly. Furthermore, its implementation into a regular OR has to be well thought-through and practical so as not to take up valuable operating space and not to place a financial burden on the hospital or organization employing it. Furthermore, the use of the technology needs to be effortless so as not to disrupt daily clinical practice and heighten the pressure on surgeons and other OR staff. In addition, these systems need to be adjustable and adaptable to different surgical procedures.

The data thus collected might be used to describe the single surgical work-steps and the whole surgical procedures, with the aim of modeling the overall intervention, including all possible or all necessary information. In this regard, the term “modeling” can be used to support one or more of the following tasks: “to describe”, “to understand”, “to explain”, to optimize”, “to learn”, “to teach”, or “to automate”. The issuing models can be used to compare different types of interventions, such as approaches using different surgical tools or instruments, to compare single surgeons or whole OR teams, or to evaluate and compare different surgical procedures.

Technical motivation

From a technical point of view, SPMs can be used for different aims. One of these aims is requirements analyses and product specification. The advantage for engineers is the possible graphical representation of the results instead of a representation as text. Furthermore, such data are quantifiable, thus allowing for a derivation of technical or clinical constraints as, for instance, average durations of performance steps, total number of performance steps, or number of iteration cycles. Additionally, benchmark studies can be performed to compare surgical strategies, substitutive technologies, or approaches or to assess the skill level of resident surgeons in comparison to senior surgeons. Product comparisons are another use-case for SPMs. By means of these comparisons, for instance, the invasiveness of one surgical product can be evaluated in contrast to another product. In addition, the data can be used for risk and quality management. Furthermore, the SPMs can be used as input for workflow management systems so that the infrastructure and technical systems in the OR can be automatically triggered or controlled according to the actual surgical situation. These and other technical merits will be further described in “Model application examples”.

Terms and definitions

To define an SPM, various other notions have to be defined first, such as medicine, surgery, process, and model. However, the next paragraphs will only give a very brief introduction into this topic and not a comprehensive account, as this is not within the scope of this paper.

“Medicine” is the “science or practice of the diagnosis, treatment, or prevention of disease” [19]. As a branch of medicine, the overall goal of “surgery” is the same. To achieve this goal, surgery is using special means. Thus, surgery is a branch of medicine treating diseases and conditions in humans or animals using manual or instrumental procedures to diagnose, cure, or alleviate a pathological condition [19], [20], [21]. The term Medicine 4.0 is adapted from the definition of Industry 4.0 [22]. It emphasizes strategies and technologies for patient-specific individualization of treatment, hybridization of medical treatment and accompanied services, and the increasing involvement of stakeholders of the surgical treatment, such as hospital administration.

A “work-step” or activity” can be defined as “a description of a piece of work that forms one logical step within a process” [23]. In consequence, a “process” can be defined as a “network of activities and their relationships, criteria to indicate the start and termination of the process, and the information about the individual activities, such as participants” [23]. In accordance with this definition, a “surgical process” can be defined as “network of surgical or surgery-related activities and their relationships, criteria to indicate the start and termination of the process, and the information about the individual activities”, whereas a “surgical work-step” or “activity” can be defined as “description of a piece of work that forms one logical step within a surgical process”.

Furthermore, a “model” is a simplified and constrained description of a fraction of reality that is built for a specific purpose. Thus, “modeling” can be understood as a simplified (formal or semiformal) representation of a suitable excerpt of reality using a suitable method for representation [24]. In consequence, an SPM can be defined as “a simplified (formal or semiformal) representation of a network of surgical or surgery-related activities and their relationships, criteria to indicate the start and termination of the process, and the information about the individual activities”. It is important to bear in mind that the simplified representation should reflect a predefined subset of interest [25] and that the method of representation should be suitable to the overall application of the model.

Delimitation

SPMs focus on the description of surgical work-steps. These activities represent the most important value-adding processes within the OR. Therefore, the goal of SPM needs to be to suitably represent these value-adding processes. Within an SPM, every surgical process needs to be represented as a chain or a network of single surgical or surgery-related work-steps (such as device setup in the OR, patient positioning, briefing, or instrument passing), including detailed information.

In most cases, SPMs cover the cut-suture-time of a surgical procedure. Activities before or after this period are usually not considered within SPMs. In general, SPMs focus on intraoperative work-steps. Perioperative events, such as anesthesiologic activities or activities performed by nonmedical staff within the OR, are of little or no importance.

However, there are some exceptions in which some of these external activities are of interest for the surgical workflow. This, for instance, is the case for a quantitative assessment of the impact of an SAS on the overall intervention course. In this case, the setup times of these systems, which occur preoperatively, need to be taken into consideration.

Modeling strategies

In general, there are two different strategies for modeling surgical processes: top-down and bottom-up strategies.

Model types

Model information can be expressed by symbolic or numeric values. This section introduces both main types briefly. However, it is also possible to combine both model types. Top-down modeling results in symbolic model types, and bottom-up modeling can result in both types.

Data acquisition strategies for bottom-up modeling

There are different existing strategies for data acquisition. They all have in common that there is no “one and only” universal strategy to encompass the multitude of potentially interesting entities within the OR. In the following section, different data acquisition strategies will be described. As per definition, these strategies are not needed for top-down modeling strategies.

Data acquisition by observation

Data acquisition strategies are methods that are of interest to the process modeler and can be employed for the realization of surgical and technical use-cases. In [25], [26], a quick and flexible strategy for the acquisition of data produced by different entities in the OR by means of observation has been described. This strategy is based on the definition mentioned above, namely, that a surgical process is a succession of surgical work-steps [27], [28], [29]. These work-steps are in turn represented by various entities that are represented as perspectives. These perspectives are, for instance,

• The functional perspective describing “what” is done in a surgical work-step,

• The organizational perspective that describes “who” is performing a work-step,

• The operational perspective that describes the “instrument(s)” used to perform a certain work-step,

• The spatial perspective that represents “where” a workplace is performed (e.g. at which anatomical structure), and

• The behavioral perspective that describes the “time” at which a certain work-step is performed.

Actual data acquisition is performed by means of tablet PCs that are equipped with a specialized software operated by especially trained observers (cf. Figure 1) [30]. The observers are present in the OR during the complete course of the intervention and record every surgical work-step they observe.

Figure 1:

Workflow editor for recording SPMs (cataract surgery example configuration) [26].

This data acquisition strategy has been validated and it has been shown that the mean accuracy for the description of surgical work-step contents was approximately 92% and the temporal delay less than 2 s [25]. The benefit of this strategy is that it was unnecessary to include a work-step of knowledge abstraction in the processing pipeline that interprets numerical data into clinically interpretable knowledge, is usually surgery specific, and might be very complex.

Using this data acquisition strategy, more than 1000 single surgical processes have been recorded, thus demonstrating that the method can be used independently for different surgical disciplines. The recorded interventions were from ENT-surgery, cardiovascular surgery, neurosurgical surgery, pediatric surgery, and interventional radiology. The models thus acquired were of varying complexity according to the length and intricacy of the surgical process recorded. For instance, median model sizes for cataract surgeries comprised 20 activities. Other, more complex interventions comprised 100–150 activities (spinal disc surgery or sinus surgery) or 200–250 activities (mitral valve repair). The most complex interventions recorded, namely, interventions for brain tumor removal, comprised more than 400 single work-steps.

Model generalization and transformation strategies

The ultimate goal for the developers of SPMs is the creation of models that can, according to the paradigm of individualization in Medicine 4.0, flexibly represent a multitude of surgical cases, disciplines, patients, and instruments. However, this aim is hard to achieve due to some characteristics of surgical processes, the most challenging being the high complexity and the high variability: adverse or unexpected turnouts of events are possible at every stage of the intervention, causing the overall workflow to deviate from the expected “standard”. This variability depends on many aspects, such as patient or surgeon properties, employed technical resources, and the hospital. Additionally, these interventions can be unstandardized as well as unplanned and otherwise highly variable.

To be able to encompass this high level of variability of surgical processes, approaches using top-down models, on the one hand, need to detail each possible outcome in sufficiently high resolution. Button-up model approaches, on the other hand, need to find strategies to transpose valid single recordings to represent other cases as well. This can be achieved by means of model generalization.

The basic idea of the generalization of workflow recordings is to merge activities. For an improved distinction, the terms patient-individual SPM (iSPM) will be used for an SPM before generalization and generic SPMs (gSPMs) will be used for an SPM after generalization subsequently.

The single process steps first need to be identified by defining certain features, such as, for instance, recognizing equal work-steps in different processes (cf. Figure 2). These work-steps can be used for process mining, a strategy transferred from business information systems. Thus, after the selection of the feature in focus, the various elements from the SPMs are merged and transferred to represent a new work-step in the generalized model. If a feature is unique, i.e. if it does not appear in any other recorded workflow, it is not merged and results in a new branch of the generalized SPM [53].

Figure 2:

Simplified gSPM generation procedure example with surgical activities A …C and statistical transition probabilities.

This strategy can be supported by an application of sophisticated data mining methods [54]. The choice of these methods is dependent on the objective and the size of the actual study. It can be assumed that the higher the number of merged individual process models is, the higher and the more stable the resulting generalized process model will be. Figure 3 shows an example of a gSPM for activities of a surgeon while performing different surgical activities during cataract procedures.

Figure 3:

gSPMs for Capsulorhexis phase in cataract interventions (solid line style, “mean” path of right hand; dotted line style, “mean” path of left hand).

Nodes indicate activity descriptions and average performance time of the activity, and edge label indicate local (upper value, the predecessor-successor node probability related to the predecessor) and global transition probabilities (lower value, the predecessor-successor node probability related to the overall iSPM sample). The gSPM was filtered by removing edges with global transition probabilities of less than 5% to remove blur.

In the example shown in Figure 3, 53 iSPMs of inpatient cataract surgeries were recorded and merged into a gSPM. The iSPMs containing surgical work-steps of the left and right hands of the surgeon were measured individually. Subsequently, activities of the same name in the iSPMs were combined into one activity in the gSPM, such as all activities of “paracentesis (with) paracentesis knife (with) right hand (at) cornea” (see bottom activity in Figure 3). If activities of the same name are followed by different activities, this results into a branching of the model. This approach enables different strategies: on the one hand, the gSPM contains all variability that was measured between the iSPMs; on the other hand, each iSPM is still present in the overall model and can be tracked to the gSPM.

Modeling languages

Basically, there is no designated or recommended language for the modeling of SPMs yet. Which language is chosen therefore entirely depends on the given circumstances, such as the requirements on the system, the scope of use of the models, the choice of system itself, or the expressive powers or the mightiness of the language.

In addition to the minimum requirement to be able to represent nodes and edges, the available elements of the language determine whether or not a language is applicable within a certain approach. Possible characteristics are the degree of mathematical formalization, the representation of spatial or temporal information, and the representation of probabilities but also the coverage of qualitative information such as the detection of conditions or risks associated with single work-steps.

A language with a high level of formalization should be employed whenever the process models to be represented need to be used as procedure algorithm for computer programs. The advantage of a language with a high formalization level furthermore is the possibility of verifying the models as early as during the design phase, for instance, with regard to the accessibility of work-steps or the presence of blind ends.

In general, there is a multitude of languages in existence that could be used to represent SPMs. Overall, a number of more than 650 process modeling languages are known. Most of these languages have their origin in information systems. The most influential and best-known of these languages for the field of business information systems are the XML Process Definition Language (XPDL) created by the Workflow Management Coalition (WFMC), the Business Process Modeling Notation (BPMN) and the Unified Modeling Language (UML) both developed by the Object Management Group (OMG), and the Business Process Execution Language (BPEL) created by the Organization for the Advancement of Structured Information Standards (OASIS). Other languages include the Event-Driven Process Chains (EPC) or Yet Another Workflow Language (YAWL). Languages with a higher level of formalization include Petri nets, state charts, and activity charts. A review and evaluation of modeling languages for SPM can be found in [55].

SPMs for visualization and analysis

Alongside the performance of analyses, SPM is one of the easiest yet most effective application areas to allow for a useful and utilizable visualization of the course of a surgical intervention. In the past, it has become apparent that these visualized procedure courses are a useful support instrument for interdisciplinary work, such as a cooperation between surgeons and engineers. By means of this visualization, both professions can gain a thorough understanding of the single work-steps and the overall procedure, thus allowing them to combine their professional points of view to find trigger points for optimizations or to discuss the impact of new SAS.

The use of SPMs to the advantage of clinical users is a necessary criterion for an effective application. Moreover, the applicability across different intervention types and diverse disciplines of surgery is a key requirement. In the following section, the implementation of SPMs will be demonstrated in several use-cases.

Evaluation and comparison of surgical approaches

In [58], SPMs were used to optimize patient treatment strategies. In this study, a potential advanced training for surgeons from an ophthalmological department of a university medical center was planned based on the models generated from their performed interventions, called surgeon-individual treatment profiles (SiTPs). The surgical interventions observed were standard cataract procedures. These were performed by three experienced surgeons during their daily routine. A total of 105 procedures were recorded and computed used to generate iSPMs. For each surgeon, separately, the resulting iSPMs were again computed into gSPMs (cf. Figure 4). Surgeon-individual workflows, activity frequencies, and median performance durations of activities were analyzed and compared. By comparing the three SiTPs, concrete recommendations for further training could be derived. Especially for gSPM segments with extended performance times or an increased number of iterations of surgical work-steps, the respective surgeon might benefit from an individualized and specific training.

Figure 4:

SiTPs as gSPMs of three surgeons for the intervention phase Capsulorhexis.

The graph shows the most commonly recorded work-steps for surgeons “red”, “blue”, and “green”. The activities are labeled in the nodes and the edges are labeled with global predecessor-successor probabilities. The gSPM was filtered to remove transitions with low probabilities to remove blurring.

SPMs for workflow management in the OR

In modern ORs, the need for intelligent information presentation has been identified to prevent an information overload for surgeons and other OR staff [60], [61], [62], [63]. In the pertaining literature, context-dependent support is frequently mentioned as one of the chief objectives in the research field of SPMs [64], [65], [66], [67], [68]. However, these systems need to be intelligible and not overtaxing for their user(s) [69] as well as suitably balanced between proactivity and transparency [70].

Proactive support is based on appropriate SPMs. In [67], [71], the design and implementation of a surgical workflow management system based on SPMs that can provide a robust guidance for surgical activities was described. The number of required iSPMs was investigated by a randomized selection of iSPMs to create a gSPM and subsequent testing of a disjunct iSPM against the gSPM. Additionally, the method of investigating the number of SPMs needed to develop such a system was investigated. The models were mapped onto workflow nets, a dialect of Petri nets, as workflow execution templates to outline the behavior of the intraoperative workflow management system. The application of SPMs for the implementation of situation-dependent assistance is expected to reduce the workload of the surgeon during endoscopically performed ENT procedures as described in [72], [73], [74].

There are many possible application fields of surgical workflow management systems within the scope of Medicine 4.0 in the digital OR of the future with integrated medical device networks. These include programmed parameterization and governing of SAS, a situation- and context-dependent information visualization for surgeons, resulting, for instance, in a timely presentation of previously acquired patient examination results or the providing of decision support systems for resident surgeons. Additionally, the ensuing surgical workflow management system could be linked to the hospital information system (HIS), thus allowing for an automatic and timely call for the next patient according to the predicted beginning of the next intervention. These use-cases could ameliorate the surgical process qualitatively and therefore be beneficial to the patients’ safety.

The developing integration of modern medical devices and IT systems in the ORs of the future is an empowering technology. A wide variety of implementations of SAS – from management and administration to support – may be constructed based on data provided by medical devices, such as online prediction of the remaining procedure time [75] or estimating the probability of medical device applications [68].