Random and systematic medication errors in routine clinical practice: a multicentre study of infusions, using acetylcysteine as an example

Abstract

Aims

The nature and incidence of errors in prescribing and giving medicines have previously been estimated by trained observers, or by retrospective analysis of incidents in which patients have come to harm. We have examined prospectively in routine clinical practice the concentrations of intravenous infusions of a drug (acetylcysteine) which is given according to a complicated dosing schedule.

Methods

We prospectively collected samples before and, where possible, after the infusion of acetylcysteine in 66 anonymous patients requiring treatment for acetaminophen (paracetamol) overdose in four centres in the United Kingdom. We measured the concentration in each infusion bag, and deduced from the weight of the patient the percentage of the anticipated dose that had actually been given.

Results

The experimentally determined dose was within 10% of the anticipated dose in 68 of 184 individual bags (37%), and within 20% of the anticipated dose in 112 bags (61%). Doses in 17 bags were more than 50% from the anticipated doses. In three patients, values in all three bags appeared to be systematically wrong by 50% or more; in a further seven cases, individual bags differed by 50% or more from the anticipated value. The median difference between pre- and post-infusion samples was 0% [interquartile range −5.2% to +14.6%], but 9% showed a disparity of greater than ± 50%.

Conclusions

Our data suggest that there is large random variation in administered dosage of intravenous infusions. Systematic calculation errors occur in about 5% [95% confidence interval 2, 8%] of cases, and major errors in drawing up in a further 3% [1, 7%], with inadequate mixing in 9% [4, 14%]. While we have no evidence that patients were adversely affected, and while the regime of administration of the drug studied (acetylcysteine) is complicated, these data suggest that the delivered dose often deviates from the intended dose, and that methods of quality control are needed.

Introduction

Health care is not as safe as it should be. Medical errors are a significant cause of morbidity and mortality [1–3]. In the United States it is estimated that 100 000 deaths each year occur as a result of medication errors (twice the number of deaths attributable to motor accidents) [4]. Unfortunately despite their importance it has proved difficult to obtain reliable data on the incidence and prevalence of medication errors.

Many previous studies have been observational. For example, pharmacists have recorded errors found by inspecting drug charts, or have watched the preparation of injections. There is always uncertainty about whether an error, detected before administration, would have occurred in practice. Covert observational studies may still miss important errors, or detect apparent errors that would not have led to harm. Other studies have relied on anonymous reporting of harmful or potentially harmful incidents [5]. These investigations are helpful in defining points in the system at which errors can occur, but give a very incomplete picture of the true incidence of errors. All miss some errors. For example, blunders in making up solutions may not be detected by inspection of the drug chart, or by direct observation, and may not be recorded by spontaneous reporting systems because they remain undiscovered.

We wished to circumvent some of the difficulties of previous studies, and also to distinguish between random variation in making up solutions and systematic errors in calculation. We used, as a paradigm [6], acetylcysteine infusion. Treatment with intravenous acetylcysteine is commonly used in the United Kingdom in patients with acetaminophen (paracetamol) poisoning. The dose depends on the patient's body weight and requires the preparation of at least three separate infusion bags [7]. The dose of acetylcysteine in each bag is different, as is the volume of 5% glucose solution in each infusion (see Table 1). Acetylcysteine is stable in glucose solution, and relatively easily measured [8].

Table 1.

Recommended regimen for the administration of acetylcysteine for paracetamol poisoning. The bags are given consecutively.

Bag number	Dose of acetylcysteine required	Volume of glucose 5%to be mixed with	Time to be given over
1	150 mg kg−1	200 ml	15 min
2	50 mg kg−1	500 ml	4 h
3	100 mg kg−1	1000 ml	16 h

Methods

We conducted a four-centre prospective study of the concentrations of acetylcysteine in the first, second, and third infusions for a series of patients treated with the standard acetylcysteine regime. Patients, prescribers, and others involved, were anonymous. The doses prescribed were based on the patients' weight, and either calculated directly or read from a table by a junior doctor. These values were then usually further checked by either a member of the nursing staff or pharmacist. Experienced nursing staff prepared infusions in three of the centres, and pharmacists in the fourth. We aimed to collect at least 150 samples (50 sets of three samples) in the 3 month period August–October 1999. We expected this number of samples to allow us to be 95% certain of detecting a difference of 25% or greater between the concentration in one bag and another in the same sequence, if it occurred in 2% or more of sequences.

All centres used ‘Parvolex’ brand of acetylcysteine (Medeva Pharma Ltd, Leatherhead, Surrey KT22 7PQ, England), and glucose for infusion from Baxter Healthcare Ltd, Thetford, Norfolk IP24 3SE.

At the beginning of administration of each infusion bag, 5 ml of infusion fluid was drawn into a sample container and labelled with the patient's weight, the centre identity, and the number (1, 2, or 3) in the sequence of infusions. Samples were then stored frozen at −20 ° C before transmission in batches to a central laboratory (Llandough Hospital) for analysis. Pilot studies had shown that mixing of acetylcysteine with 5% glucose solution is rapid, but to check that this was so, participating centres were asked, where administratively possible, to send a second sample from each bag at the end of infusion. We did not expect this to be possible in every case, because of movement of patients, administration of the whole volume before sampling, and changes in staff on duty.

Acetylcysteine concentration was measured by an established method [8]. Samples were freshly prepared before injection and analysed in duplicate in batches of 30. All samples from one patient were assayed together. A five-point calibration curve prepared from pharmaceutical acetylcysteine (Parvolex®) at 20, 40, 60, 80, 100 mg ml⁻¹ was constructed for each batch. Quality control samples were made up in a 5% glucose solution at concentrations of 25 and 50 mg ml⁻¹ and included in each batch. Inter-assay and intra-assay coefficients of variation were, respectively, 6.8% and 2.9% at 25 mg ml⁻¹, and 3.9% and 2.1% at 50 mg ml⁻¹.

Results were analysed as percentages of the anticipated concentration of acetylcysteine, obtained by converting raw data, expressed as concentrations in mg l⁻¹, into percentage of the concentration for the specified bag, based on the weight of the patient, the sequence number of the bag, and the recommended dosing schedule.

We allowed for the fact that manufacturers overfill infusion bags, with target volumes of 274 ml, 535.5 ml, and 1049 ml for 250 ml, 500 ml, and 1000 ml bags, respectively [Personal communication from the manufacturer]. To verify the manufacturers' assurance that bags were made to narrow tolerances, and that the contents remained stable over time, we weighed bags that nominally contained 1000 ml glucose 5% solution with different expiry dates, 12, 17, and 21 months ahead. Mean filled weights in grams were 1113.4 (n = 5, s.d. 4.9); 1115.2 (n = 4, s.d. 3.4) and 1114.4 (n = 5, s.d. 1.0). Empty weights were similarly consistent: 53.8 g (n = 10, s.d. 2.6).

We also allowed for the small differences in protocols between centres – in one centre, 50 ml was removed from every 250 ml bag before the addition of acetylcysteine, and in another, a volume of glucose solution equal to the calculated volume of acetylcysteine was removed before addition of the acetylcysteine. Protocols, and materials used, were otherwise the same, except that in one centre, pharmacists prepared infusions.

We expressed the results for each patient in whom we had all three pre-infusion samples as mean, X_bar, weighted by the proportion of total dose in each bag, and range (greatest value–smallest value). We also examined the upper and lower control limits (UCL_Xbar and LCL_Xbar) overall and for each of the centres. UCL_Xbar and LCL_Xbar are approximations to the upper and lower bounds of the 99% confidence interval for X_bar. Such methods are widely used in statistical process control [9, 10]. We also examined the differences between preinfusion and postinfusion samples, and the values for all infusion bags.

Results

We received 184 separate samples, including 57 complete sets of three infusion bags. Duplicate samples were obtained at the end of infusion for 128 bags, including 35 of 57 complete sets of initial samples.

The distribution of X_bar, the weighted mean dose as a percentage of the anticipated dose for each of the 57 complete sets of samples is shown in Figure 1, with UCL_Xbar and LCL_Xbar. The experimentally determined dose was within 10% of the anticipated dose in 68 of 184 individual bags (37%), and within 20% of the anticipated dose in 112 bags (61%). Doses in 17 bags (9%) differed by more than 50% from the anticipated doses.

Figure 1.

The frequency histogram for per cent anticipated acetylcysteine concentration in each of the bags from the four centres combined demonstrating the median and interquartile range.

The number of sets whose weighted mean dose was between 90% and 110% of the anticipated value was 24/57 (42%), and between 80% and 120%, 41/57 (72%). One patient received 157%, 239% and 205% of the anticipated dose; one 167%, 169%, and 177%; and one 51%, 68%, and 57%.

In 7/184 bags, values of 305%, 239%, 215%, 215%, 205%, 54%, 52%, and in 1/184, 0% of the anticipated dose, were recorded.

On average, patients received 107% of the recommended dosage [LCL_Xbar=76%, UCL_Xbar=138%]. The results for each of the four centres were (X_bar, [LCL_R, UCL_R]): 101[78, 124], 144[89, 199], 97[79, 115] and 91 [60, 120]% of recommended dosage. In three centres, the average range between highest and lowest bag in a set was less than 30%, but in the fourth centre, the average range was 54% [LCL_R=0%, UCL_R=139%], making the average range overall 31% [LCL_R=0%, UCL_R=80%].

The median difference between samples taken before and after infusion was 0% [interquartile range of −5.2% to + 14.6%]. Twelve of 128 samples (9.4%) showed a disparity of greater than ± 50%. One preinfusion sample contained no acetylcysteine, and the corresponding postinfusion sample contained 77% of the expected amount.

Discussion

The processes of calculating, drawing up, sampling, analysing, and extrapolating the anticipated and administered doses are all subject to error. We have estimated the random variability in the process and major error rates associated with different components of the process. We have assumed that, as intended, the samples were taken from a consecutive series of patients, or at least, that the sampling was not biased. If there were biases, they would have been likely to minimize the errors we detected – for example, by excluding patients who received treatment when centres were busy.

We considered a difference of 50% or more between measured and anticipated values to indicate major error in calculation or preparation, not random variation, and the desirable difference to be 10% or less. Prior to the study, we thought most patients would receive within about 10% of the anticipated dose, but only one third did; and one third differed by more than 20% from the anticipated dose. In 3/57 sets, values in all three bags appeared to be systematically wrong.

We took a substantial difference (greater than ± 50%) between preinfusion and postinfusion samples to signify inadequate mixing, and that was the case in over 9% of the infusion bags examined. Failure of mixing, which could expose patients to higher concentrations than intended during a specific part of that infusion, might contribute to a dose-dependent adverse effect such as histamine release. Such effects are commonly seen with acetylcysteine infusions.

Degradation of acetylcysteine in storage, and deviation of the product concentration from specification, are unlikely, but we have not examined them explicitly. Assay errors are probably small compared with the errors in other parts of the process, since samples were assayed in duplicate with quality control standards. Inaccuracies most likely arise in adding the correct amount of acetylcysteine solution to the correct volume of 5% glucose solution: syringes may contain air bubbles; acetylcysteine ampoules that nominally contain 10 ml are incorrectly filled and when charts of dosage against weight are used, then weight may be rounded to the nearest 2.5 kg.

In three cases, major errors seem to have occurred in calculation with all three infusion bags being consistently inaccurate. We cannot say whether the calculation errors were made when prescribing or when drawing up. One or both processes require a conversion from units of weight to units of volume before the acetylcysteine solution is drawn into a syringe.

The isolated errors where the concentration of acetylcysteine in a single bag in a set of three differed by a factor of 2 or 3 from the expected value seem more likely to be due to errors in drawing up than in prescription, since the prescriber is able to check the three prescribed doses for internal consistency. One bag in which no acetylcysteine was found in the preinfusion sample probably represents a protocol error in which the sample was taken before the addition of acetylcysteine, since the postinfusion sample contained acetylcysteine.

We estimate the chances of a significant error in prescribing calculations as 3/57 (5%); of a major error in drawing up as 7/184 (3%); and of a major error in the order of sampling as 1/184 (0.5%). Around 9% of bags are inadequately mixed (Table 2). These compare with industrial error rates of 3% for ‘errors in simple arithmetic with self-checking’ [11].

Table 2.

Rates of major errors in different parts of the process of prescribing and giving acetylcysteine^a.

Operation	Rate of major error	Approximate 95% confidence intervals
Calculation	3/57 (5%)	2.3–8.2%
Drawing up acetylcysteine and adding it to glucose	7/184 (3%)	1.0–6.6%
Mixing	12/128 (9%)	4.3–14.4%
Sampling	1/184 (0.5%)	0.0–1.6%
Analysis	assumed zero

^aMajor errors are assumed when values differ by more than 50% from anticipated values, or when the difference between two samples (pre and postinfusion) is more than 50%.

The process described may overestimate the amount of error occurring in the preparation of most intravenous infusions as the preparation and administration of acetylcysteine is especially complex. Complex systems provide multiple opportunities for error. Methods to try to prevent such errors occurring include education, having adequate and standardized protocols that are relevant and easy to follow and regular re-validation of the procedure being performed [12]. However all these methods continue to rely heavily on the individual performing the task; the overall method involved should also be examined to see if it can be simplified and if further safety margins can be introduced into the system. For example, instead of giving three bags at three different concentrations, and three different volumes being required, it may be possible to administer the dose in two bags of a similar volume. Modern technology, such as computerized dispensing, might help to reduce the amount of error occurring.

Computer-assisted prescribing can reduce error rates [13], and might substantially reduce computational errors. When errors occur at a later stage in the process of drug preparation or administration, computer-assisted prescribing cannot easily prevent them, and they cannot be reliably detected by previously described methods of studying medication errors.

Medication errors have not, to our knowledge, previously been assessed by the standard methods of industrial quality control.

Not all medication errors translate into patient harm, and acetylcysteine is a relatively safe drug. However, errors of the magnitude we detected here, where half or twice the intended dose of an intravenous drug was administered to some patients, would have serious effects for drugs with a low therapeutic index. Drugs such as insulin, inotropic agents, phenytoin and fosphenytoin, aminophylline, diamorphine, and morphine, all of which are used in emergency care, represent the most serious potential dangers.

We now have the tantalizing prospect that, with simplified analytical methods for acetylcysteine, and possibly other drugs, continuous methods of quality control could be introduced and allow us to be sure that patients receive what we intend them to receive. Unless better systems can be introduced to ensure that consistent standards are maintained, our study suggests that potentially serious problems with dosing errors will persist, even with perfect systems for prescribing.