Simulation-Based Reorganization and Automation Planning in Clinical Laboratories

Parts of this publication were presented at the LabAutomation conference in San Diego, Jan 21, 1998

In industry, computer simulation as a subdiscipline of operations research, has been used for many years to support complex decisions such as robotization of processes or planning new factories. Clinical laboratories and diagnostic manufacturers having recently been faced with the challenges of total laboratory automation and factory type re-engineering of procedures are just about to learn, how powerful this tool can also be to for their purposes. It helps to reduce the risk of failure and to assess the impact of expensive equipment prior to making buying decisions (1,2).

The basis of our experiments is a computer program called Simlab® V 3.0, which we originally developed from 1992 to 1994 (1,2). This DOS-based PC program was written in Simscript II. In 1996 and 1997 we added new modules to the original program, which now allow us:

To facilitate the tedious process of workflow data collection

To present the simulation output data in numerical and graphical formates

To export Simlab data to commercial spread-sheets such as MS Excel.

A technical article describing these new developments has been submitted for publication (3).

Input and output data of Simlab® V 3.0 are summarized in table 1.

Table 1:

Input and output data of Simlab® V 3.0 (for details see ref. 1)

Static Input Data (Laboratory Design)	Dynamic Input Data (Laboratory Workflow)	Output Data (Performance)
Work areas (accessioning, clinical chemistry, etc.)	Hourly workload pattern (number of orders): Random arrival rates of corresponding samples are created by Simlab	Utilization of instruments
Staff (function, working hours, etc.) and assigned instruments (type, throughput, down-times etc.).		Utilization of staff
Analytical tests (names, assigned materials and instruments) and corresponding sample tubes (primary, secondary)	tests or test panels ordered for each sample tube	Sample queue lengths
Work rules (sample routing, priority handling)	Priorities (routine or STAT) for each test	turn-around times (overall and per work area)

Typical procedure of workflow simulation

A simulation study for a workflow analysis in a laboratory can be broken down into four phases (typical duration in parentheses):

• A:

Experimental design phase (1–2 days)

B:

Computer modelling (2–3 days)

C:

Experiments (2–4 days)

D:

Report and presentation (about 3 days)

The procedure includes the following 10 steps (phases in parentheses):

• 1)

Analysis of the current laboratory design and workflow (A)

2)

Definition of goals and design of computer experiments (A)

3)

Creation of data input files (B)

4)

Establishing a “virtual laboratory” model on a PC (B)

5)

Validation of the model, i.e. comparison of the “virtual performance” with reality (B)

6)

Conducting what-if-experiments (C)

7)

Assessment and plausibility check of experimental results (C)

8)

Financial assessment of promising scenarios (C)

9)

Report writing (D)

10)

Presentation of the report, making decisions (D)

In general, steps 1, 2 and 7–10 can be done by the laboratories or industry clients themselves, whereas steps 3–6 require the involvement of a specialized workflow analyst. In those cases, where the study does not involve a real laboratory, e.g. industry studies for market segments or general laboratory types, step 5 can eliminated. Instead, a result validation with experts is recommended.

In this section we will present and discuss typical results of two studies conducted for clinical laboratories. Part of these studies was presented at international venues (4–6).

Pilot study at the University of Virginia in Charlottesville

This study was conducted in 1996 in order to test the performance of Simlab under real-life conditions. The study led us to the decision to develop the input data file generator and the data presentation module depicted in fig. 1. These tools allowed us to greatly simplify the entry of daily workloads, so that we were able to accomplish the study within 1 week and present it to the audience of an international workshop (4) immediately after the last experiment. The workload pattern, generated with the new input file generator, Simlab V 3.0 creates random arrival rates from this file thus eliminating the need for an LIS interface or for typing data in manually.

Fig. 1:

Typical output from the data presentation module.

Two major results of the study are:

Given the current workload of the laboratory, the workload of the technicians in the accessioning area would be decreased considerably

The IDS frontend would be reasonably utilized

Our data indicate that the laboratory would definitely have a benefit from front-end automation in two respects: Personel savings and a decrease in total turnaround times from 150 min to 93 min (data not shown).

Reorganization study at the German Heart Center in Munich

This study was conducted in 1997 to support planning for future expansion of the laboratory. The laboratory had just moved to a new building with higher capacity for heart operations. So the experimental design was: What if the sample workload of the laboratory increased by factors of 1.5, 2 or 3, and how would the laboratory cope with this increase without hiring too many additional people. The most dramatic effect of doubling the workload is on turnaround times. The pattern is shifted from a peak around 45 min to a variety of new peaks between two and six hours. Since analytical speed is the most important goal of a heart surgery department, the major goal of the reorganization scenarios described below, was to restore the original turnaround time.

Examples of how doubling the number of arriving samples would lead to delayed and quite unpredictable workload peaks in the accessioning area and to unacceptably high staff utilization. Furthermore, samples will queue up especially in the area of manual aliquoting for three different coagulation analyzers and bottle-necks in the coagulation area will be “masked” due to the preceeding delays.

Fig. 2:

Effects of doubling the number of arriving samples on workload peaks and staff utilization in the accessioning area. This simulation eperiment was performed at the German Heart Center in Munich.

We then experimentally improved the “chaotic” situation of scenario 1 by hiring people and/or installing appropriate equipment. The major goal of these experiments was to restore the orginal total turnaround time of about 1 hr. We found that this could be achieved by:

Hiring 4 people (senario 2)

Hiring 3 people and installing a Tecan Genesis aliquoter (scenario 3)

Hiring 2 people and installing a Behring BCS coagulation instrument (scenario 4)

Hiring 1 person and installing both instruments (scenario 5).

All of these four scenarios were designed in a way that they resulted in acceptable turnaround times of about one hour. Fixed costs per tests were highest in scenario 2 and lowest in scenario 5, due to the fact that salaries have higher impacts on fixed costs than the proposed equipment.

Fig. 3:

Experimental evaluation of various optimization scenarios in terms of turnaround times and fixed costs per test.

This means that the laboratory should choose the combination of automated aliquoting (e.g. Tecan Genesis) and consolidated coagulation testing (e.g. Behring BCS).