Our study results show a significant shift in guideline-concurrent AB type after receiving guidance from the virtual mentor. This represents a move forward in terms of responsible AB prescribing, as guidelines encourage the practice of prescribing narrower spectrum SAPs, which contributes less to AMR.

The results of this limited 3-hospital pilot suggest that the use of a mobile phone app can have an impact on the SAP prescribing patterns of busy surgeons. In our study we showed the potential for an app to significantly change prescription behaviours of surgeons, including shortening the recommended course-length, and becoming more compliant with standard guidelines.

Our results show the power of relatively simple AMS and educational tools, similar to a recent Cochrane review [21], in which the implementation of an antimicrobial stewardship programme was linked to a decrease in the length of the antimicrobial course.

However, the implementation of AMS programmes can be challenging. Especially in resource-limited settings like our study sites in Nigeria. There is a need to adapted AB prescribing guidelines to the local antimicrobial resistance profile; however, these data are often lacking [22]. By focusing on key interventions like decreasing the length of the AB course or working towards the elimination of using AB with overlapping spectra substantial effects can be made in resource-limited facilities with few to no infectious diseases specialists [23, 24].

In a recent systematic review form the WHO about overuse of medications in LMICs, it was estimated that the overuse of AB ranged between 18.4 and 97.0% [23].

In our pilot, 25% of AB were prescribed without clear indication. There are many reasons for prescribing antibiotics against standardized guidelines [20]. There is a lack of oversight, as antimicrobial stewardship programmes in Nigerian hospitals are inadequate [26] and if present, guidelines are being weakly enforced [27]. Additionally, there are factors such as cost and availability [22], social pressures from patients and senior colleagues, and a low inclination from physicians to follow policies [28].

Antibiotic overuse is driven by multiple factors, not limited to doctors prescribing without indications. We can also look to the use of overly broad or redundant combinations of multiple ABs. Unnecessary AB combinations, often used for gram positive and anaerobic bacteria, account for a large portion of AB use [29]. In our pilot project surgical patients were prescribed ABs for SAP which were too broad in their coverage. A 1st generation Cephalosporin provides a narrow spectrum and targeted use for common skin bacteria [18] and represents a safe choice in regard to efficacy, risk of side effects and risk of developing resistance [30,31,32]. In our pilot we found that though 86 (25%) surgeries required only a 1st generation Cephalosporin, even after feedback its singular usage was a rare choice by the surgeons.

There is no singular reason for overprescribing; one can divide the myriad reasons into individual level drivers which are clinician or patient driven, and system-level drivers which are institutions and organizations driven. Examples of individual level drivers include patient wishes and physician knowledge about AB use as well as the local resistance patterns. System based drivers include staffing, resources, and national and local guidelines [25]. In 42% of cases, surgeons in our pilot reported ‘following local practices’ as the reason for guidance discordance, while ‘patient’s environment’ requires prolonged dose in 21% of cases [20]. In both cases, the listed reasons for guideline discordance included both individual and system level-drivers.

There are two distinct moments that are critical for training the health workforce to prescribe antibiotics responsibly: during training in medical school, and during continuing-education training for working professionals. There are training gaps in both instances, and they need to be considered separately in order to be adequately addressed. Based on the surgery risk, type and duration results of our study, the GADSA app is a prime example of an intervention that could be best suited to the iterative, daily needs of a working professional.

As part of surgeon training, sensitivity towards patient environment needs to be taught, and inclusion of this element needs to be included on a systemic (curricular) level. For example, surgeons need to be trained to consider patients living in remote area with no possibility of post-surgery follow-up. Surgeons who are aware of patient constraints are able to use existing guidelines and better tailor the AB course to ensure compliance and suitability.

One major strength of the app is the inclusion of prescribing surgeons and pharmacists in the design and decision-making process of the app. Taking local prescribing patterns and habits into account, will likely enhance adherence to new guidelines and empower physicians in their knowledge of following standardized guidelines and preserving their patients individual safety [23].

Another strength of the app is the unique, gamified approach to AMS. We were able to engage busy surgeons by incorporating little challenges and digital prizes [16, 17]. With a mobile-decision support tool like GADSA, health care professionals in resource-limited settings have ready access to up-to-date information via their smart phones, that is easily adaptable to local settings and in addition functions as an educational tool [33].

Limitations of the study include the researcher’s access to sensitive prescribing records, a lack of complete, electronic records, and the ability to ground—truth the existing records with patient reality. A lack of electronic medical record system makes information sharing a challenge. Pre-intervention paper data about AB used was collected by a pharmacist and transcribed into a digital format. As such, only prescriptions filled pre-surgery by the patient in the hospital pharmacy were collected. We were therefore unable to compare the data collected, pre-intervention by hand and post-intervention through the app.

Post-intervention, we are also not able to ascertain, that surgeons actually prescribed their recorded AB, instead of reporting a socially acceptable answer in the app but prescribing something else in reality. However, as we had maintained a close relationship and trust with the surgeons via the WhatsApp groups and the local PIs, this self-reporting bias was most probably minimized.

While the WhatsApp group was not designed to change behaviour and is not part of the GADSA app, one cannot exclude that the surgeons were influenced in their activities and behaviour through interaction with each other and the researchers. In a scoping review of the role of WhatsApp in medical education, Coleman and O’Connor found, that the use of WhatsApp may be effective as a learning tool [34]. Given our WhatsApp group was not designed with an educational programme and no clinical cases were discussed, it certainly helped retain and motivate surgeons to use the app, the influence on their prescribing behaviour however was likely small. An implementation of WhatsApp as an integral part of the GADSA App is a possibility for future research.

Given the lack of antimicrobial resistance data of the hospitals, the guidelines used were internationally standardized and recognized guidelines [18, 19]. With locally adapted guidelines, surgeons might have a stronger trust and therefore conviction to follow these guidelines.

Furthermore, we could not rule out that in some cases the junior surgeons entered the data on behalf of the senior surgeons. Thus, in the next development, a multi-user version will be designed to address this challenge.