Automation of Aseptic Sterile Preparation: Risk Analysis and Productivity Comparison with Manual Process

Manual production controlled area

Three isolators with 4 workstations were used, one for cytotoxics, one for monoclonal antibodies and one for “on demand” tailor-made cytotoxics with 2 workstations. Isolators were ISO 5 (class A) closed enclosure and were located in an ISO 8 (class D) cleanroom in conformity with French Good Manufacturing Practices (GMPs) for hospital and community pharmacies [11]. When possible, additionally to the per-process visual control, each preparation was subjected to analytical control of the end-product using HPLC-UV detection method or a spectrophotometer UV-Raman (QCRx, Icônes Services, Sucé sur Erdre, France).

Semi-automatic production controlled area

One isolator (class A) located in ISO 8 (grade D) cleanroom with 2 workstations were available, and one of them was equipped with a peristaltic pump Verderflex^® Vantage 3000 (Verder, Eragny-sur-oise, France) (Figure 6). This pump allowed us to compound standardized dose batches from bulk solution either for cytotoxic or anti-infectious drugs. The Verderflex^® pump was a peristaltic roller pump system designed for filling bags, syringes, diffusors from drugs and liquid solutions [12]. The pump design was as a one-piece bloc with a numeric keypad on the front panel for controlling and regulating the pump. On the front of the pump, a housing contains the rotating roller, which ensures the transfer of the liquid when the tubing is installed. When the bulk big bag is connected to the tubing, it allows the division into a defined number of smaller bags of end-product with a programmable flow and volume. The PcPump^® software allows to run the filling process of the bulk bags and end-product bags. A foot pedal is located under the isolator and allows starting or stopping the filling process. The peristaltic pump was chosen according to its compatibility with peracetic acid gazing sterilization process used to enter the isolator through the pass-through. An analytical control with HPLC-UV is performed on the bulk solution to check the final concentration. Then each bag is individually weighted to check the conformity with the expected weight of the final bag end-product.

Robotic production controlled area

The KIRO^® Oncology robot (Kiro Grifols, Arrasate, Spain) was brought through an “innovation market”, and enabled to compound sterile drugs including cytotoxics, monoclonal antibodies and anti-infectious drugs. The robot was inside an ISO 5 (grade A) BioSafety Cabinet (BSC) IIb allowing manufacturing sterile injectable preparations. According to the process (closed transfer process) the BSC was located in an ISO 7 (grade C) cleanroom validated in accordance with hospital GMPs [11].

It was fitted with 2 mechanical arms running in parallel to perform different tasks at the same time. Right arm holds vials in order to reconstitute them using 2 peristaltic pumps while left arm holds syringes in order to withdraw the expected dose (Figure 7). The cabinet was only opened for loading and unloading whereas it was closed during the entire compounding and self-cleaning processes allowing protecting operators and environment from exposure to hazardous drugs as well as embedded bins located underneath each arm. The robot was equipped of a self-cleaning process of the enclosure which is run at the end of each working day.

It provided an in-process qualitative and quantitative control allowing checking each step. Qualitative control is ensured by barcode and data matrix readers and cameras located inside the cabinet and in the user area allowing to identify each vial, bag, syringe or others devices used. Furthermore, a datamatrix code is assigned to each material or device entering in the robot allowing to guarantee traceability until preparation release. Quantitative control is provided by a gravimetric method weighting each vial, bag and syringe at each step during the compounding process by internal scale and weighing each final preparation with external scale.

According to the supplier, the KIRO^® Oncology system is compatible with many types of final containers (bags, syringes, cassettes and elastomeric pumps) marketed by numerous providers. It allows to use ready to use solutions and powder to be reconstituted. It is able to do on demand tailor-made preparations or standard preparations following a “series production process” (i. e. repetition of the same preparation of same dose in the same final container), but it was currently unsuitable for batch production from bulk solution which will need specific development for both mechanistic and software. A daily process of production includes starting operations (i. e. clothing, manual cleaning, loading of raw material into cleanroom, microbiological sampling), production phase itself including robot cycles and the daily final cleaning (manual cleaning followed by self-cleaning). Cycles were divided in pre-processing phase (preparation of the required material and loading of the robot), automated process in robot and post-processing phase (unloading of the robot, manual external weighing and pharmaceutical validation if necessary).

Manual process

This process was the reference process for the preparations of monoclonal antibodies (bevacizumab, nivolumab, infliximab) produced “on demand” tailor-made. A junior pharmacist (internship) who was not involved in the production collected the data. This process was performed using isolator technology, a sterilization (or disinfection process) was required before entering the isolator. For each preparation, a tray is prepared and combines all materials needed for one preparation (i. e. drug, diluent preparation devices). The time for loading sterilization hatch was divided by the number of contained trays. The sterilization time spent to gas sterilize the hatch workload was not considered as additional time because it was performed simultaneously to other tasks. Similarly, the time for the analytical analysis with analytical device (HPLC-UV or Raman-UV) was not considered. The manual process was based on one operator preparing and a half-time assistant operator for visual in-process controls on 7.5 hour working day without lunch breaktime.

Semi-automatic process – VERDERFLEX^® pump

Three batch productions of an injectable anti-infectious drug (acyclovir) to be reconstituted at one single dose were performed. As previously, the sterilization time and the analysis time by the analytical control device were excluded. The data were collected by the junior pharmacist. The working campaign for one batch production of 50 bags with the pump was based on 1.5 half-time operators.

Robotic process – KIRO^® Oncology robot

For this process, real life “on demand” tailor-made production days of monoclonal antibodies and anti-infectious drugs (bevacizumab, nivolumab, infliximab, acyclovir) were observed by a junior pharmacist, as well as two simulated days of dose standardized preparations (acyclovir) and anticipated preparations (cytotoxics). The first day was performed using single drug, prepared from powder to be reconstituted (acyclovir), at different standardized final doses. The second was simulating cytotoxic multi-drugs preparations, prepared from ready-to-use solutions (irinotecan, 5-fluorouracile, gemcitabine and oxaliplatin) at different final standardized doses. Subsequently, two additional days of dose standardized, and anticipated preparations were achieved. These additional days corresponded to two single doses, one prepared from ready-to-use solutions (5-fluorouracile) and one prepared from powder to be reconstituted (cyclophosphamide). This last was defined as a mixed production. The working day was based on 7.5 hours based on a single operator running the robot fulltime (during lunch break time, the operator was switched with another one) and included pre- and post-processing tasks (i. e. clothing, manual cleaning) without any break during lunchtime.

Statistical analysis

Descriptive statistics (mean, standard deviation (SD)) and simple analyses (Student test) were conducted. A p-value ≤ 0.01 was considered statistically significant.

Choice of risk analysis method and definition of the team

In accordance with our quality department, we have decided to implement the FMECA method as an analysis tool for our risk mapping. It is indeed the most suitable for the objectives set and especially the most widespread within several sectors of our university hospital. It is a reference method in the medical and hospital world [13]. It was recommended by the Institute Healthcare Improvement (IHI) in United States [8, 9, 14, 15] and by the “Haute Autorité de Santé” (HAS) in France [16]. The major interest of this method is a quantification carried by the criticality index (CI). This takes into account three important parameters in the occurrence of the risk: the frequency, the severity of the effect if it occurs and the possibility of detection before the event occurs. These elements are particularly useful in deciding which steps in the process need to be improved or modified.

We formed a multi-disciplinary team consisting of pharmacy technicians, health manager, senior pharmacists, junior pharmacists assigned to injectable drugs preparation. Once the group was formed, the study managers trained it in the FMECA methodology.

Realization of the risk analysis

Risk analysis was based on gross criticality approach to determine general criticality to the mentioned risk. Weekly sessions of approximately 45 minutes were planned, a total of 9 sessions were held. The first session was dedicated to train the working group, the n°2 and n°3 focused on the manual process, a session was needed for the robotic process, and finally three sessions for the semi-automatic process. During these sessions, the group agreed on the chronological sequence of the work steps of each process. Then for each step, in brainstorming session, we asked ourselves “what could go wrong?”. Each failure mode was rated using 3 rating parameters (Occurrence (O), severity (S) and detectability(D) by using a consensus risk-rating grid validated by our pharmacy quality department (Figure 1) for the institution. The multiplication of the 3 gave the critically index (CI) of each step. We allowed to use a restricted scoring grid to simplify the analysis and to be in accordance to our institution policies. The use of a restricted scoring grid limited the CI score scale but did not modify the conclusions, as far as the ranking of CI is the point to be considered in FMEA method more than a value. At the end of each session, an Excel^® tool was completed by the junior pharmacist who was one of the project pilots. Then the failure modes were classified according to their CI. During the last session, the most critical ones were exposed to the working group in order to find means of controlling the risk or solutions, and to make them acceptable. Then, each plan was examined with the pharmacists in charge of the production unit to refine its relevance, feasibility, pilots and achievable deadlines. In addition, the root causes of the failure modes were determined using a 5 M diagram. Starting from the identified effect, the analysis is conducted around 5 categories:

1. Environment: workstation, physical organization

2. Methods: procedures, information flows

3. Means/Materials: equipment, machines, tools

4. Manpower: human resources, staff qualifications

5. Material: consumables, raw materials

Figure 1:

Risk scoring grid.

Finally, a session was planned to present to the entire production unit team, the action plan drawn up and action plan monitoring indicators selected.

Process mapping

Before analyzing each process, the working group agreed on the chronological sequence of the work steps. Each process was divided into 3 main steps: pre-process, per-process and post-process. The manual process was divided in 20 sub-steps, in 26 for the semi-automatic process and in 18 for the robotic process.

Failure mode distribution by process step and determination of the risk priority rating (RPN)

For the “what could get worse” question, the group identified 68 failure modes for the manual process, 79 for the semi-automatic process and 42 for the robotic process. The sum of CI was 1060 for the manual process, 719 for the semi-automatic process and 656 for the robotic process. For each failure mode, the CI is the product of the O x S x D scores. Theoretically, the minimal CI of one failure mode was CI_min=1×1×1=1 (O=1, S=1, D=1) and the maximal CI of one failure mode was CI_max=4×5×4=80 (O=4, S=5, D=4). In comparison to the manual process, the pump permitted to decrease of 32.2 % the sum of CI and to 38.1 % with the robot. The ratio number of failures mode on number of sub-steps was 3.2; 3.0 and 2.3; for manual, semi-automatic and robotic process respectively.

The number of failure modes according to process is given in Figure 2. We identified a number of failure above the mean CI (i. e. 16) at 22, 27 and 13; for manual, semi-automatic and robotic process respectively. The Risk Priority Number (RPN) was set at 36 to allow the analyze of failures for the three processes, a higher RPN would had excluded robotic process. The selected RPN allowed us to identify 11 priority failure modes. Ishikawa diagram of failure modes is presented on Figure 3, and a synthesis of their CI are presented in Table 3. Additionally, eight low risk failure modes were considered specifically for the semi-automatic pump which accumulated lots of failures modes in comparison to other processes but mainly with a rate under 36. Considering the 3 processes together, 87.7 % of the causes of the failure modes were mainly the result to a “Manpower” issue, 12.7 % about a “Means/Materials” issue, 10.6 % about a “Environment” issue, 5.8 % about a “Material” issue and 3.2 % about a “Method” issue. “Manpower” was always the first cause. “Environment” was the second cause for the manual process. In semi-automatic and robotic, “Means/Materials” was the second cause. Their distributions are described in Figure 4. Examples cited by the working group are presented in Table 4.

Figure 2:

Number of failures modes according to the process.

Figure 3:

Ishikawa diagram of priority failure modes.

Figure 4:

Distribution of per-process failure causes by categories.

In common with all processes, the human errors were the most involved in the analysis of cause identified. In this category, forgetfulness, lack of training, tiredness and lack of time were most often mentioned by the working group. However, human factor was less important for the robotic process as it was independent of human for most of the preparation steps. On counterpart, with robotic or pump, the technical issues were the most identified. For manual process, apart of human error risks, working environment was the second cause identified i. e. lack of space or unsuitable organization.

Establishing preventive measures

Throughout the session 8, the work group worked on these 11 priority failure modes to find solutions and set up means of controlling the risk. Others were considered like acceptable or low priority because the risk was controlled. The reduction of the criticality index was based on two actions: either the reduction of the failure frequency that requires a change of organization, or the increase of the detectability of the failure by the implementation of a control. Examples of action plans are given in the Table 5. This last meeting session allowed a collegial and pragmatic reflection on the action plan. Finally, 19 failure modes were analyzed, and 31 actions were planned and monitoring indicators were built accordingly. The indicators for monitoring the actions planned were divided in common actions and specific actions related to the 3 different processes. In common actions, we identified education improvement, review and creation of quality documents and environmental microbiological and chemical monitoring. Related to automated processes, technical issues were the main issue to be followed. Among common indicators, one can cite:

1. education monitoring indicators e. g. number of education program per year, number of operators educated per year. Number of operators passing a post-education exam (results>80 % of good answers). The ratio between the number of quality documents read by operators and the total number of quality document to be read.

2. ratio between the number of quality documents written/ or revised and the total number of documents to be written/ or revised.

3. Quantitative and qualitative collection and analysis of non-conformities.

Examples of actions planned and monitoring indicators thereof are given in Table 5.

Then, each action with the help of monitoring indicator were planned to be periodically reviewed to refine its relevance, feasibility, pilots and achievable deadlines with the pharmacists in charge of the production unit.