Exploring the extent of post-analytical errors, with a focus on transcription errors – an intervention within the VIPVIZA study

Introduction

Quality in laboratory medicine has previously been defined as “an unfinished journey”, and it is still a vital quest to continue the journey, as laboratory information is of increasing value for physicians for making appropriate clinical decisions [1]. High-quality laboratory information is also essential for patient safety, as it has been shown that 19 % of laboratory errors lead to further inappropriate investigation, and 6.4 % of laboratory errors were associated with improper care [2].

Laboratory errors were previously synonymous with the test’s analytical performance. However, at present, the concept of laboratory errors has, hand in hand with technological developments and increased awareness, developed into a different meaning. It is now evident that the total testing process (TTP) should be considered to enhance the quality of laboratory information [3]. The TTP can be divided into three phases: pre-analytical, analytical, and post-analytical. It is estimated that the error frequency in the pre-analytical phase ranges from 46 to 68.2 %, 7–13.3 % in the analytical phase, and 18.5–47 % in the post-analytical phase [4], 5]. Thus, the pre- and post-analytical phases are more error-prone [6].

The pre-analytical phase and its errors correspond to all steps, from preparing the patient to collecting and transporting the sample to and within the laboratory. This has previously been described in detail [7]. Analytical errors originate from within the laboratory. Over the last decades, extensive work with quality assurance and within the laboratory has already been done. Therefore, there has been an impressive reduction in analytical errors [8]. This has been thanks to the widespread use of internal and external controls and the formation of laboratory networks such as the Cholesterol Reference Method Laboratory Network (CRMLN) to develop standardisation protocols to ensure the accuracy of the analytical methods [9]. Additionally, the implementation of laboratory accreditation has further improved the development of the quality assurance journey [10]. Post-analytical errors include, for example, incorrect transcriptions, reporting and interpretation of lab results. The literature is sparse in general for the extra-analytical phases, but for the post-analytical phase, the literature is exceptionally scarce since it is difficult to establish their frequency [11]. However, the post-analytical phase is increasingly recognised as an essential step in maximising the quality of laboratory information [12].

The extent of errors in the post-analytical phase varies since the reported error frequencies depend on the method used to detect them. Even though there have been improvements in recent years, e.g. by requirements for post-analytical processes according to the ISO15189 standard [13], there are not yet any widely accepted standardised methods to detect the errors in the extra-analytical phases available, leading to different methods for estimating error frequency in different studies [14], [15], [16], [17]. This further contributes to the uncertainty of the true error frequency, especially for minor errors [18].

Whilst transcription errors have been reduced due to digitalisation in laboratory medicine, they have not been eliminated, and there are no automated methods to detect them. Additionally, many transcriptions are still performed in other areas of clinical practice, e.g. when using different risk prediction algorithms. SCORE2 is a risk prediction algorithm for cardiovascular disease (CVD) prevention. It is based on traditional CVD risk factors such as age, smoking status, systolic blood pressure, and non-HDL cholesterol [19]. Depending on the level of risk factors, the patient will be assigned to a risk category (low-moderate, high or very high). In clinical practice, the physician often transcribes the lab results and risk factors on a website to calculate the risk category and recommend a suitable preventive intervention to the patient.

Hence, we aimed to investigate the post-analytical error frequency in the VIPVIZA trial, focusing on transcription errors in lipid parameters at two different time points. We also aimed to examine how the errors affected the SCORE2 risk prediction algorithm for cardiovascular disease, which includes non-HDL cholesterol, as this could lead to an incorrect treatment decision. Additionally, we wanted to determine whether education among the research personnel between the time points decreased the extent of errors.

Materials and methods

Results

The number of transcription errors identified in VIPVIZA 2019 and VIPVIZA 2023 study can be seen in Table 1. As four lipid parameters were assessed (TC, LDL-C, HDL-C, and TG) for every individual, there were 12,460 transcriptions in VIPVIZA 2019 and 10,576 in VIPVIZA 2023. Some individuals had more than one error in the lipid parameters. Consequently, in VIPVIZA 2019, there were 69 transcription errors, affecting 56 individuals. This error frequency corresponds to 1.80 %, and when converted into parts per million (ppm) to 18,000. In 2023, the corresponding number of transcription errors was 26, and the number of individuals with errors was 25, corresponding to 0.95 % (9,500 ppm). The frequency of errors approximately halved from 0.55 % (5,500 ppm) in 2019 to 0.25 % (2,500 ppm) in 2023.

The number of individuals that were misclassified according to the SCORE2 risk prediction algorithm are visualised in Figure 1 for VIPVIZA 2019. As non-HDL cholesterol is the sole lipid parameter included in SCORE2, individuals with transcription errors in total and/or HDL cholesterol were the only ones affected, corresponding to 39 individuals. In total, we identified six individuals that would have been risk misclassified, i.e. 15 % of the transcription errors resulted in risk misclassification. Of these six individuals, three were assigned a lower risk category after correction, and three were assigned a higher risk category. The distribution of individuals in different risk categories can be further seen in Figure 1. In the 2023 VIPVIZA survey, none of the individuals had a transcription error in total or HDL cholesterol. Consequently, no risk misclassifications were identified at this time point.

Figure 1:

The extent of transcription errors leading to risk reclassification based on the SCORE2 category in VIPVIZA 2019.

Discussion

We found that the frequency of transcription errors was 0.55 % (5,500 when converted to ppm) in the 2019 VIPVIZA research database, with 1.80 % (18,000 ppm) of individuals having one or more errors. The extent of these errors halved after the educational intervention, resulting in a transcription error frequency of 0.25 % (2,500 ppm) and 0.95 % (9,500 ppm) of individuals in 2023. Thus, it was observed that this type of intervention was effective in a controlled environment, such as among research personnel. We also discovered that the transcription errors led to SCORE2 risk misclassification in 15 % of cases, potentially resulting in patients receiving an incorrect risk assessment or physicians making incorrect treatment decisions.

This study is an example of a simple, applicable method for data quality assurance in research databases, which is an essential aspect of publishing high-quality research with generalisability. To our knowledge, this error identification method of lipid parameters has not been previously published. A strength is that the method is easily reproduced and identifies post-analytical errors of a smaller magnitude. Errors of smaller magnitude have been a challenge to identify in the research field, but also a well-known problem in the extra-analytical phase within the laboratory community, concluding that the actual error frequency is under-detected [4], 18], 22]. For instance, a previous study found that the true incidence of errors was at least four times the actual events detected [18].

The true error frequency is highly dependent on the method for error detection, and currently, no uniform method for managing errors exists for the post-analytical phase. For example, the error frequency previously reported was based on the clinicians being asked to pay extra attention to suspected laboratory errors and discuss them with a laboratory physician daily [2], 15]. Other described methods are that the laboratory personnel double-check the results [16] or identify an edit in the computerised system [17]. Additionally, many previous studies reporting post-analytical error frequencies are old since laboratories report digitalised results and are currently not dependent on transcription. This makes it a challenge to find studies with updated frequencies of the true error frequency that represents the current situation, where most transcription occur outside the laboratory environment, e.g. in point-of-care-systems, The reported error frequency in the post-analytical phase has been estimated to be 18.5–47 % and thus varies greatly depending on the method [4]. Of the studies that have explicitly reported transcription errors, the error frequency is estimated to be 11.7 % in one study [16] and 1.1 % in a study from a transfusion laboratory [17]. However, many studies on post-analytical errors have not assessed transcription errors explicitly and, therefore, not reported the extent. Similar studies in fields other than the laboratory community have, however, assessed transcription errors, such as when administering pharmaceuticals. The reported transcription error frequency in a study assessing medical errors in cancer patients was estimated to be 11.54 % [23] and 2.21–8.11 % in an intensive care unit with transcriptions of blood glucose levels [24]. Thus, the errors are considerably more common in clinical practice compared to our study, estimated to be 0.25–0.55 %.

In our case, the extent of errors was assessed as part of a pragmatic trial, and it is not unreasonable to hypothesise that the error frequency is higher in clinical practice because of the high number of staff involved, the need to cut expenses, and work under stress, compared to the research environment. However, our reported error frequency is quite similar in magnitude to the one reported in a transfusion laboratory, which also reported a reduction of errors after educational intervention among the personnel, consistent with our results [17]. This may indicate that the laboratory community’s extensive quality journey has increased the awareness of potential sources of errors and, therefore, reduced their error frequencies compared to other medical fields. Still, many transcriptions are conducted in clinical practice daily, such as entering laboratory results into risk motors or results from point-of-care systems.

Despite our relatively small error frequency, we still estimated that 15 % of the individuals with an error in total and/or HDL cholesterol would have been risk misclassified according to SCORE2. If the same magnitude of transcription errors are assumed to be present in clinical practice, this would lead to several patients receiving incorrect pharmacological treatment and risk assessments. Moreover, transcribing all laboratory variables and other risk factors into a risk motor is common, further yielding a potential for more transcription errors than indicated by this study. In addition, we also noticed that in the VIPVIZA 2019, there were eight transcription errors in LDL-C, and three of these participants would have missed out on receiving lipid-lowering therapy according to CVD prevention guidelines because the transcription error presented a falsely low LDL-C value. Thus, the post-analytical quality of laboratory results is an essential aspect of patient safety.

Regarding the SCORE2 risk misclassification, there are some crucial aspects to consider associated with the known limitations of the risk algorithm [25]. Because the risk algorithm assesses the absolute risk for a cardiovascular event in the next 10 years, younger individuals with a high burden of risk factors will still be assessed as low risk, as long as the risk factor burden is not extreme, i.e. an underestimation of the actual CVD risk is common in young individuals. In this study, this means that participants could have a significant transcription error in lipid variables (>1 mmol/L). However, despite this, the participant is still not classified into a higher or lower risk category if the other risk factors are normal, and the individual is young. Nevertheless, the transcription error could have been clinically relevant in other aspects.

There are some limitations to this study. For example, there is still a possibility that our method for error identification does not identify particularly small errors<0.1 mmol/L; consequently, there is a chance that the true error frequency is, in fact, higher than reported in this study. However, these minor errors correspond to a magnitude similar to the analytical uncertainty and would not affect the clinical assessment. There is also the possibility that minor errors are present in other variables in our research database, as these (e.g. blood pressure) are also transcribed by the research nurses, but we had no standardised method to detect them. Consequently, we did not account for the fact that other risk factors included in the SCORE2 algorithm (blood pressure or smoking status) probably have transcription errors, leading to an underestimation of the risk misclassification frequency reported in this study. Additionally, as this study only assessed the error frequency within the research database, it was not possible to assess whether the errors led to actual clinical events, i.e. if the participants received incorrect treatment or not, as previously presented in other studies investigating the consequences of laboratory errors [1].

Nevertheless, this study sheds light on the importance of evaluating the quality of laboratory information in research databases and adds further updated information regarding the true frequency of transcription errors. Additionally, we illustrate the potentially harmful effects of entering incorrect lipid values into a risk motor such as SCORE2. The reported error frequency might be modest, but we still highlight the detrimental effects of transcription errors that might lead to incorrect risk assessments and treatment decisions in individual cases, especially as we hypothesise that the error frequency is higher in clinical practice. We therefore emphasise the further need for digitalisation to assure patient safety.