Medical prescriptions are often the last encounters patients have with healthcare practitioners. Their writing is regulated by professional guidelines and the law.^1,2,3,4 Prescription-related inaccuracies may impact negatively the quality of care and compromise patient safety, and they may lead to medication errors that have significant social, financial and legal consequences.^5,6,7,8 Such errors can arise from errors related to medication prescribing, dispensing, administration or patient compliance;^9,10 however, the origin of majority of errors noted in the medical prescription can be traced to writing by healthcare professionals,¹¹ which can then have knock-on effects such as delayed dispensing.¹² Prescription-writing errors account for up to 70% of medication errors that could potentially result in adverse effects.¹³

A total of 412 prescriptions were analysed across all districts, with each of the hospital sites contributing approximately 25% of the entire sample. Table 2 is a descriptive account of the specific characteristics for each district hospital OPD as well as for the four district hospitals combined. Significant differences were found between the hospitals with regards to the average number of prescriptions per day, the number of prescribers on the day and the prescribers’ professional category mix. The average number of prescriptions per day varied between 100 and over 300, and that of prescribers varied between six and eight on the days studied. Medical officers constituted 83.25% of the prescribers overall. Adult patients accounted for 96.12% of all prescriptions written.

Figure 1 illustrates the full completion in percentages of a prescription’s 19 essential elements across the 412 prescriptions assessed in Gauteng South; the raw count and percentages showing fully completed, partially completed and not completed elements of the prescription for Gauteng South and for each of the four district hospitals are included as Appendix 3. It is worth noting that one prescription was signed for by a student and not counter-signed by a medical officer, and existing internal memos allow clinical associates and medical interns to write OPD prescriptions, notwithstanding the challenge with regard to the legal status of clinical associates as authorised prescribers, as this category has not been accommodated in the schedules as required by the Medicines and Related Substances Act.

Prescribers achieved rates of 90% and over for completion of the date of the prescription, the treatment regimen, the prescriber’s signature and handwriting legibility. The patient’s file number was indicated in 83% of prescriptions. Lower rates were seen for the use of generic names of medications (39%), abbreviations and decimal points (34%) and for other components of patients and prescribers’ identifiers. Particularly, low completion rates were seen for patients’ sex (25%), age (34%) and weight (2%).

Table 3 is a compilation of GASs: 37.52% of prescriptions scored 80% – 100%, whilst 62.38% scored 40% – 79%. The mean score out of 38 was 27.92 (73.38%) ± 5.53 s.d.

The inability to achieve a desired GAS of ≥ 80% was noted for the following elements: patient identifiers (24.51%), the use of a medication’s generic name (38.83%). the use of recommended abbreviations and decimal points (34.70%) and prescriber identifiers (58.50%). High GAS values of ≥ 80% were noted for the following prescription elements: date of prescription (99.75%), treatment regimen (99.51%) and legibility (90.05).

The distribution of the GAS categories across the district hospitals in Southern Gauteng is indicated in Figure 2. Only in Ekurhuleni did more than half (59%) of the audited prescriptions achieve a score of 80% – 100%.

We examined possible associations between specific characteristics of each OPD and prescription-writing GASs, which are summarised in Table 4. Significant negative associations were found between the ability to achieve a GAS of 80% – 100% and the number of OPD prescriptions per day (p = 0.001) and the number of OPD prescribers on the day (p = 0.005). No associations were found between the prescription type or prescriber category and adherence to prescription-writing guidelines.

Seeking to understand the effects of the number of prescriptions written per day as well as of the variability in the number of providers on GAS values, we conducted logistic regression analysis, which indicated that the number of OPD prescriptions per day and the number of OPD prescribers on the day were significant predictors of GAS values (Table 5).

Significantly, the hospital with more than 300 OPD prescriptions per day was 70% less likely to score 80% – 100% compared to the hospital with 100–200 prescriptions per day (odds ratio [OR]: 0.30; 95% CI: 0.14–0.62).

The relationship between the number of OPD prescribers (continuous variable) and GAS indicates that an increase in the number of prescribers on a day was associated with an odds ratio of 0.73 in the likelihood of scoring between 80% and 100% (OR: 0.73). In other words, with every increase in the number of prescribers, there is a reduced likelihood of scoring between 80% and 100%, or reaching the target goal. Although this seems counterintuitive in light of the argument that high workload increases the likelihood of a reduced GAS, it suggests, using advanced statistics beyond the scope of this study, that it is possible to calculate optimal numbers of OPD prescriptions and OPD prescribers on any given day.

To the best of our knowledge, this is the first effort in South Africa to develop a single tool that combines domestic regulatory and international best practice models to audit prescription-writing accuracy in the country’s primary care settings. The results of this study provide a baseline understanding of challenges with prescribers’ adherence to prescription-writing regulations. The results also suggest some possible reasons for omissions and inaccuracies in prescription writing that could guide future quality improvement initiatives.

While sampling was restricted to four district hospitals’ OPDs in Southern Gauteng, the required sample size for data collection was exceeded (374 intended; 412 actual), lending strength to the results. It was also apparent that not all the sites were in fact the same to start with, with statistically significant differences between the sites in relation to average number of OPD prescriptions per day, average number of providers and the professional category of prescribers (Table 2). These differences persisted in terms of GASs across the four sites, with interesting variations.

The composite GAS indicates that only seven (1.69%) of the 412 prescriptions assessed achieved the desired 100% or gold standard, and only 155 prescriptions (37.52%) scored 80% – 100%. In Bertha Gxowa and Kopanong district hospitals, 59% and 44% of the prescriptions achieved the GAS of 80% – 100%,respectively, compared to 16% and 31% in South Rand and Yusuf Dadoo district hospitals, respectively. This is a significant difference (Kruskal–Wallis: p = 0.001), largely attributed to the higher scores achieved for completion of patient and prescriber identifiers which, if not counted, would bring all the four district hospitals to similar lower GASs (Kruskal–Wallis: p = 0.9).

Out of 38 total possible points, the mean score for all 412 prescriptions was 27.92 (73.38%) ± 5.53 s.d. In Ife, Nigeria, Erhun et al.¹⁴ reported global percentages of 1.3% and 85.9% of completely filled medical prescriptions, respectively, from a health centre and a teaching hospital, and in France, Fourgon et al.²⁵ reported 17% of fully accurate ambulatory patient prescriptions. Similar to our settings, most of these deficiencies pertained to omission errors.

The main omissions found in this study related to patient identifiers and prescriber identifiers in 56% and 23% of prescriptions, respectively. This may be explained by the practice of prescriptions being attached to the patient’s file which already contains all of the patient’s particulars, including the address, telephone contacts and medical diagnostic information. The prescriber in a busy OPD would rarely take the time to re-transcribe these details from the file to the prescription, and it seems this omission also occurs when it comes to write his or her own identifiers.

Practices reported in the literature to enable improved completion of these required identifiers include the consistent use of a sticker or label with the patient’s details (with the birth date rather than the age) and the mandatory acquisition and consistent use of self-inking stamps with the required prescriber identifiers with only a signature required on the prescription.^23,24 These would be low-cost improvement measures as the sticker or label printing infrastructures already exist in the OPDs and self-inking stamps are not expensive.

High prescriber adherence to regulatory guidelines included completion of the treatment regimen (99%). This is possibly because of the dispensers who, as the last gate keepers, created an institutional culture that places higher emphasis on the treatment regimen writing aspect of the prescription. This is in contrast to other settings where omissions of various components of the treatment regimen have been reported, with some even up to 91% incompleteness.³¹

Similarly, the date of the prescription was present 99% of the time; this was possibly because it is almost always stamped on the prescription chart by the clerk at the time of the patient’s registration. In contrast, the date of prescription was omitted in up to 18% of prescriptions in the reported literature.²⁶

Of note in this study is the fact that the generic names of medications were not used for 61% of prescriptions, and non-recommended abbreviations and decimal points were used for 65% of prescriptions. Similarly, high proportions of poor compliance in this regard are reported in the literature.^19,29 The Human Sciences Research Council in South Africa found in 2005 that amongst general practitioners only 11% prescribed by generic name and generic prescribing was below 50% in public facilities.⁵⁰ Education and individual prescribers’ efforts can improve on this at little cost.²¹

As has been evidenced elsewhere, the adoption of an electronic health record with electronic prescribing that includes standardised fields with prompts for certain patient conditions may result in greatly improved accuracy in all prescription-writing elements considered in this study. This may appear to be a high tech or high cost measure, but may be worth the investment given its positive impact on quality and reducing medication error,^44,45,46 especially in our environment with increasing malpractice and adverse event litigation.

Significant differences were noted between the district hospitals for GAS values and the essential elements of patient identifiers, prescriber identifiers and legibility. Although no association was found between the prescription type (noting the small numbers of paediatric and controlled substances prescriptions) and the prescriber’s professional category, performance differences have been reported between prescriber categories in the literature.^16,17,18,19 This also could have happened because not all categories of prescribers were present in the OPD on the day prior to data collection. Therefore, other prescribers at that facility would have been missed and potential differences never examined. Similarly, the significant difference in legibility scores may have been caused by one prescriber in a particular district hospital on that day.

The negative associations found between GASs and the number of OPD prescriptions and prescribers per day are suggestive of a negative impact of a heavy workload on the accuracy in prescription-writing, as also reported by Ajemigbitse et al. in Nigeria.¹⁹ As suggested by the logistic regression model for predicting GAS values in Table 5, it is possible to determine an optimal number of prescriptions and prescribers per day to minimise inaccuracies, and performance may be even further enhanced if e-prescribing or a computerised physician order entry (CPOE) system could be introduced,^44,45,46 notwithstanding its high cost, especially given that handwritten prescriptions are consistently found to be more prone to errors than electronic ones.¹⁸ In limited resource settings, however, prescription-writing quality improvement may still be achieved by using simple low-cost methods such as regular audits and feedback sessions, and the use of prescribers’ self-inking stamps as highlighted above.^16,17

Several limitations may restrict the generalisability of this study. The lack of diversity in prescription types was discovered only after data collection. It seems that paediatric medical records were kept in separate paediatric clinics within the OPDs and that ‘controlled substances’ records were also usually separated from other records for the purposes of oversight; moreover, as these same categories of patients attended OPD on particular days, their prescriptions may have been missed in the survey. An important additional type of prescription would be that of the elderly, which we did not examine specifically. Importantly, the ‘correctness’ of the medication as it correlated with the medical condition being treated and adherence to treatment guidelines by prescribers in their choice of medication were not assessed. Nor were prescriptions differentiated between new and repeats, or scripts for single versus multiple drugs. Also the snapshot or 1-day survey may have reduced the variability amongst the prescribers; and, to understand the impact of the numbers of patients and prescribers, the sampling mechanism could be refined. Inclusion of these aspects may enhance the results of a future study.

A combination of legal requirements and professional guidelines, drawn from local and international sources, was used in selecting variables in this study (Table 1). The legally binding requirements were drawn from the law in application at the time of data collection³ and included amongst the prescriber’s identifiers the ‘practice number’. However, ‘professional registration number’ was used as it was required as a best practice although it was not the legal requirement then. The General Regulations to the Medicines and Related Substances Act issued in 2017⁵⁰ replacing the one of 2003³ modified the legal requirement to be included in the prescriber’s identifiers from the ‘practice number’ to ‘registration number with the relevant statutory health council’, that is, the ‘professional number’ which identifies the individual prescriber better. The ‘practice number’ is not used for in-house district hospital OPD prescriptions. Its use as a variable might have had some impact on the results obtained, but not on the associations observed.

The 90% legibility score observed in this study is at odds with the literature where much higher proportions of illegible scripts are reported.^16,17,18,19 Also in the assessment of legibility, all the scoring was done by the same person, the researcher alone. His familiarity with the OPD medicines and the environments could introduce a researcher bias. However, on-site pharmacists anecdotally reported difficulty reading only less than one in 10 scripts.

This study revealed that in Southern Gauteng district hospitals, of 412 OPD medical prescriptions surveyed, one-third achieved a GAS of 80% – 100%, with two-thirds scoring between 40% and 79% for compliance with prescription-writing regulations. The most common deficiencies relate to the omission of patient and prescriber identifiers, with completion rates of 43.93% and 77.25%, respectively. Also, medication generic names and recommended abbreviations and decimal points were used in only 38.83% and 34.17% of prescriptions, respectively. Associations were found between accuracy scores and the number of OPD prescriptions per day (p = 0.001) and the number of OPD prescribers on the day (p = 0.005), pointing to a negative impact of high patient volumes and multiple providers on reducing the accuracy of prescription writing.

A prescription-writing quality improvement programme would be an appropriate framework to implement recommendations pertaining to institutions or systems and individual prescribers. These recommendations include low-cost or low-tech measures, such as the consistent availing and use of pre-printed stickers with the patient’s particulars on prescriptions and the mandatory use of self-inking stamps with the required particulars for prescribers. Reinforcement is also required regarding the consistent use of generic names of medications and being vigilant about the use of abbreviations and decimal points. E-prescribing, although a high-cost endeavour for our limited resources settings, may be considered given its positive impact elsewhere on the quality of prescription writing and medication error reduction.