| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Abstracts of Oak Ridge Posters |
1 Department of Pathology, University of Texas Medical Branch, Galveston, TX;
2 Abbott Laboratories, Abbott Park, IL
aaddress correspondence to this author at: Department of Pathology, McCullough 5.120, University of Texas Medical, 301 University Blvd., Galveston, TX 77555-0742; fax 409-772-5683, e-mail aamohamm{at}utmb.edu
Automated clinical chemistry analyzers were first introduced as single- and multichannel continuous flow analyzers in the 1960s (1)(2). Ongoing evolution of analyzers has been driven by two factors: (a) increased demand by clinicians for rapid turnaround times (TATs); and (b) a need to decrease the cost of producing laboratory results and to consolidate laboratory testing into fewer workstations. The diagnostic industry has responded to this need by developing analyzers that have high throughput and expanded test menus, and perform both clinical chemistry assays and immunoassays on a single platform. As a result of these technical innovations, analyzer throughput and assay TAT are now dependent on such factors as (a) tests ordered on samples, (b) the number of samples loaded on a analyzer, and (c) the STAT-vs-routine ratio. Selection of the most appropriate analyzer to meet laboratory TAT consistently is no longer a trivial exercise of dividing the total number of daily tests by an analyzer’s hourly throughput to determine whether it will meet laboratory needs. The goals of this study were to determine the feasibility of using MedModel 2001 (Promodel Corporation) software to develop a simulation model of the ci8200 integrated chemistry and immunoassay analyzer manufactured by Abbott Diagnostics, to estimate the percentage error in the model’s predictions of test TAT, and finally, to use the validated model to evaluate the impact of increased routine test volumes on the TAT of STAT samples.
A simulation model of a clinical analyzer uses computer software specifically designed to imitate and capture an analyzer’s dynamic behavior to study its performance under different conditions. Researchers in Europe and the United States have used computer simulations to study various aspects of clinical laboratory operation, including staff assignments, evaluating the queue length, and priority handling and processing TAT for different sample types (3)(4)(5)(6)(7)(8)(9). However, there are no published reports using computer simulation specifically to predict the minimum, maximum, and mean TATs and the test throughput of an analyzer. There are two basic types of simulation techniques: discrete event and continuous simulation. Discrete events are instantaneous actions occurring at unique points in time. Examples of discrete events include a sample arriving in the laboratory and being loaded in an analyzer rack, and the analyzer rack being loaded in an instrument. These events cause a change in system states. In this model the computer maintains a timing device (simulation clock) that advances with each event that takes place at a fixed time point. If an event represents the initiation of an activity that will conclude in the future, the simulation will add the completion time to a list of future events and advance the clock when the next event is due. Discrete event simulation uses statistical methods for generating random behavior and estimating model performance. Continuous simulation models represent actions uninterrupted over time (e.g., models simulating biochemical reactions, temperature logs, and the flow of water in rivers).
The Medmodel 2001 (Promodel Corporation) software is discrete event simulator software. Using this software, we simulated the processes that occur on the ci8200 chemistry and immunoassay analyzer (see the flow diagram in Fig. 1
). Briefly, samples are programmed to arrive at a certain sample arrival cycle. Using defined attributes, laboratory personnel identify arriving samples and randomly assign them one or more tests. The samples are then loaded on the rack, with the maximum number being determined by the holding capacity of the rack. In a typical laboratory, technologists try to optimize sample loading by waiting until the rack is completely filled or by waiting for a specified time after the arrival of the first sample. This condition is simulated by use of the code within MedModel software "LOAD 5 IN N(15,2) seconds". This means that the technologist will load 5 samples, if available, or will wait for a time period that is randomly selected between 13 and 17 s to load the rack on the analyzer. The rack with the sample is then moved to the pipetting stations. The ci8200 is a lock-step analyzer, and the time duration or sample cycle time for an individual lock step determines the theoretical throughput of the analyzer. The lock-step duration for chemistry assays and immunoassays are 4.5 and 18 s, respectively. Thus, based on the test attributes assigned to a sample, the pipettor is used for 18 s with each immunoassay test and 4.5 s with individual chemistries. The only exception is when sodium, potassium, and chloride are ordered on a sample. In this case, the pipettor is used for 4.5 s for all three tests. After the sample cycle time has elapsed, the model creates an aliquot, which is routed to an available cuvette on the appropriate chemistry or immunoassay module. Using the "create" function, we were able to attach the attributes of the mother sample to each daughter aliquot. Once all samples in a rack are pipetted, the rack is routed to its original position on a retest sample handler (RSH). When the aliquot arrives at a processor, it decreases the capacity of the processor by 1 and waits 9.72 or 29.4 min for chemistry and immunoassay tests, respectively. For electrolytes (Na+, K+, Cl–) and STAT immunoassays, the test times are 3 and 14.7 min, respectively. Once all aliquots for a sample have waited for the assigned time, they exit from the model, triggering a global variable to change states from 1 to 0 or vice versa. This change tells the software that all tests have been completed on a sample, triggering the log file to capture the TAT for the exiting sample.
|
Validation of the ci8200 model was done by running 200 routine and 20 STAT samples on 3 separate days. The test distributions for routine samples were 55% chemistry, 23% chemistry and immunoassay, and 22% immunoassays. For STAT samples, the distributions were 40%, 30%, and 30%, respectively. The following four routine and STAT sample scenarios were run each day: 200 routine; 200 routine with 10% repeats; 200 routine and 20 STAT; and 200 routine, 20 STAT, with 10% repeats. The numbers of tests run on the ci8200 for each scenario ranged from 1281 to 1434 for chemistries and 129 to 147 for immunoassays. The tests ordered on STAT samples were Chem 7 + ß-human chorionic gonadotropin (ß-hCG; 20%); Chem 7 (30%); Chem 7 + troponin I + creatine kinase-MB (CK-MB; 35%); and ß-hCG (15%). The tests ordered on routine samples were ß-hCG (4.5%); ß-hCG + Chem 7 (3.5%); troponin I + CK-MB (4%); Chem 10 (6%); health panel-20 Chemistry test panel (1.5%); Chem 4 (1%); Chem 6 (7%); Chem 7 (12%); Chem 7 + troponin I + CK-MB (8%); Chem 7 + cocaine (3%); Chem 7 + follicle-stimulating hormone (FSH) + luteinizing hormone (LH) + ß-hCG (2%); Chem 7 + liver panel (3%); Chem 7 + thyroid-stimulating hormone (TSH; 9%); Chem 12 (9%); FSH + LH + ß-hCG (1.5%); glucose (6%); liver panel (3.5%); lipid panel (3.5%); and TSH (12%). The simulated throughput was estimated by running 40 replicates of each scenario for 3 consecutive days. The 120 runs for each scenario were averaged and used for statistical calculations. The actual and simulated throughputs on the ci8200 were then compared. The percentage error between the simulated and experimental results was calculated as follows.
 |
The validated model was then used to study the impact of increasing routine test volumes on TATs of STAT samples. Using the arrival pattern of stat and routine samples at our institution, we tested the effect of increased routine test volume on TAT of STAT samples. This was done by simulating the following scenarios: 1000 routine and 320 STAT samples; 1200 routine and 320 STAT samples; 1400 routine and 320 STAT samples; 1600 routine and STAT samples; 1800 routines and STAT samples; 2000 routine and STAT samples; 2200 routine and STAT samples; and 2400 routine and 320 STAT samples. The tests ordered on these samples were same as those used in the validation study.
The experimental and simulated TAT and throughput results are compared in Table 1
. The simulated throughput was slightly less (–3.25% to –5.23%) than the actual (experimental) throughput. These differences, however, were not statistically significant (P = 0.15). A unique feature of the ci8200 chemistry and immunoassay integrated analyzer is the RSH sample handler. Whenever a STAT sample is loaded on the analyzer, the RSH "pick and place" arm picks up the sample and immediately presents it to the pipetting station. If a routine sample is already present at the station, it is preempted, thereby allowing rapid sampling of STAT samples. This could cause an increase in the routine sample TAT while maintaining a constant STAT sample TAT. As seen in Table 1
, this is indeed the case. The simulated mean TAT for routine samples increased by 98 min for chemistry and immunoassay tests when the sample load arriving over a 24-h period increased from 1000 to 2400 tubes. For STAT samples, however, the mean TAT was relatively unchanged, ranging from 12.22 to 12.52 min and from 20.06 to 20.49 min for chemistry and immunoassay tests, respectively.
|
In conclusion we found computer simulation to be a powerful tool to evaluate test throughput and TAT of a clinical analyzer. Developing a simulation models takes 3–4 weeks, and the information needed to develop a model for a clinical analyzer is usually available in the public domain or directly from representatives of various diagnostic companies. MedModel 2001 is robust software and allows a user to simulate all important functionalities of an analyzer. Although these models do not give an exact simulation of an analyzer, prediction of analyzer throughput and TAT with less than 5–10% error from actual analyzer results is possible. Numerous diagnostic companies have developed or are developing simulation models of this type to help the laboratory make an informed decision. However, before accepting the results of a simulation, laboratorians should (a) visually compare the actual sample-processing logic of the analyzer with the simulated analyzer by reducing the speed of the simulation, (b) compare the simulated TAT with actual instrument TAT to ensure that the two agree with less than 5–10% error, and (c) estimate the error in the simulated throughput vs the real throughput. We believe that in this era of "getting it right the first time", simulation studies of this nature are of the utmost importance to ensure that the chosen analyzer will meet the desired TAT with the laboratory’s specific test ordering and sample arrival pattern.
References
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
