Development of an algorithm to detect and reduce complexity of drug treatment and its technical realisation

Complexity of drug treatment arises from different characteristics in the medication regimen or external circumstances that may encumber drug administration for a patient. These complexity factors are manifold [1] and can potentially lead to nonadherence [2] and unplanned hospitalisations [3, 4]. They can be assigned to six categories depending on whether they are related to the dosage form (e.g. the use of transdermal patches [5]), the dosage scheme (e.g. an once weekly administration [6]), additional instructions concerning the use of a drug such as the coordination of the drug administration with a meal [7], the patient (e.g. swallowing difficulties [8]), the product (e.g. an intricate packaging [9]), or the medication process itself (e.g. frequently changing prescriptions [10]).

Besides consideration of single complexity factors such as the number of drugs in a medication regimen or the dosing frequency [11], the medication regimen complexity index (MRCI) is probably the most frequently used tool to assess complexity related to the medication regimen in a standardized way [12]. However, there is no tool currently available that covers all aspects potentially influencing complexity of drug treatment. Moreover, analysing and reducing the complexity of drug treatment remains time-consuming [13], even though the complexity in drug treatment could often be reduced significantly by simple measures [14]. In various health care settings, electronic tools have been developed to facilitate integration of decision aids in routine care [15] and also for reducing complexity in drug treatment it has been shown that individual complexity factors can be reduced by electronic decision aids [16]. Also the MRCI was transferred into an electronic decision aid [17] and tested in the home care setting [18]. Among patients in the intervention group, 8% of patients dropped below a MRCI score of 24.5 when the decision aid was used compared to 4.5% of patients when the decision aid was not used, suggesting that a tool that covers all areas of complexity would be even more successful.

However, there still remains the risk that patients do not adhere to the medication regimen after it has been changed, particularly if they have not been involved in the process of decision-making [19‐21] and if complexity aspects relevant for the individual patient are not specifically addressed. Hence, a potential algorithm and its technical realisation should consider patient preferences to ensure that suggested changes and trainings are tailored to the patients’ needs with maximum chance of acceptance.

The aim of this study was therefore to develop an algorithm as well as its technical realisation that comprehensively assesses and reduces complexity of drug treatment. To this end, complexity factors must be specified that could be considered in such an automated analysis, appropriate strategies to assess and integrate the patients’ perspective must be developed and integrated, and, finally, feasible suggestions to mitigate complexity must be provided.

As a prerequisite to the tool development, four key functionalities were defined:

Functionality I: Easy access to structured medication data.

Functionality II: Comprehensive, automated analysis of the drug treatment’s complexity based on distinct complexity factors.

Functionality III: Personalization of the automated analysis to the patient’s perspective and actual problems.

Functionality IV: Suggestion of appropriate measures to reduce or mitigate the identified complexity factors.

The respective functionalities were developed as follows:

The developed electronic tool comprised 52 complexity factors covering a wide range of factors contributing to treatment complexity. Thereby, almost two in three complexity factors could be automatically detected if structured medication data was entered as provided by a medication schedule. This applied in particular to the complexity factors related to the dosage form, the dosage scheme, or additional instructions, thereby confirming the results of the MRCI decision aid [12, 17].

In contrast, most complexity factors relating to the patient, the product, or the medication process could not be detected by the electronic tool, either because this information could not be deduced from data currently stored on the medication schedules or structured information to develop respective algorithms is currently not available. Hence, to integrate also these factors, either alternative data sources such as electronic health records would be needed (e.g. to have access to the patients’ diseases) and more thorough information particularly on product characteristics would be required, e.g. on the packaging of each drug or individual tablet shape and size. With such information, the medication schedule could for instance be searched for packages that are difficult to open such as child-resistant containers [9] and algorithm-based packages that are easier to open could be suggested.

Moreover, to develop more complex algorithms, e.g. for the automated detection of sound-alike drugs, either structured data on resembling drug names could be integrated or an algorithm to identify such drug pairs based on linguistic approaches, such as an analysis of orthographic and phonetic similarities of all drug names in a medication regimen, could be specified [27].

In total, about one third of the complexity factors considered relevant by the experts could not be automatically evaluated by our complexity tool. To compensate this shortcoming in the short term, we used distinct questions that complemented the automated assessment and address potential problems of the patient with the current treatment regimen. This approach proved to be sufficient for many relevant challenges in a patient’s life such as swallowing difficulties, diverse storage conditions, or complicated packaging. However, factors relating to the patients’ capabilities, their health literacy or their life style could not be assessed by a single question and hence were not considered in our assessment even though they were rated as important by the experts in the first place.

Nevertheless, the inclusion of health care professionals in the selection of the complexity factors considered in the tool and the development of the key questions ensures that the final electronic tool meets the needs of routine care. Thus, it should be made sure that the tool can be implemented easily in patient care and that relevant patient problems are addressed, which can be solved or mitigated directly by a health care professional. However, in order to reduce the influence of the professional background and the personal experiences in patient care of each individual on the final algorithm, several health care professionals with different professions or specialisations have been involved. In general, the included key questions clearly distinguish the analysis with our electronic tool from a simple complexity assessment and allow for patient-specific selection of optimisation measures (personalized intervention). The actual impact of this approach on routine care will be tested in a prospective pilot study and compared to other approaches to reduce treatment complexity that do not consider the patient preferences. A future widespread use of the tool in the health care system is conceivable because it could be used in different settings, such as pharmacies, ambulatory care and, also hospitals. Even individualization of the recommended optimisation measures to the setting appears possible; for use in a pharmacy, an algorithm to change the dosage form could for example be replaced by the recommendation for action to contact a doctor and suggest an alternative drug. Furthermore, specific patient populations might need adapted optimisation measures, such as blind or visually handicapped patients who will benefit less from standard information brochures but might need materials with verbal explanations. This work has several limitations. First, the analysis of complexity of drug treatment is based on structured medication data from the nationally standardized medication schedule and, thus, there is a risk that not all of the patient’s drugs are considered, e.g. because the medication schedule is not up-to-date or not comprehensive. In particular, pro re nata (as needed) medication and non-prescription drugs might not be included in the medication schedule, for example because they are forgotten or not considered as relevant by health care professionals and patients. However, in order to comprehensively assess and reduce treatment complexity with the electronic tool, it would be essential to include all medications used by the patient in the medication schedule.

Second, the automated identification of complexity factors is partly based on the identification of index words in the free text fields of the medication schedule. Therefore, accurate identification depends on the correctness and comprehensiveness of the medication schedule and correct spelling of the index words. Moreover the definition of index words might not be comprehensive enough to cover all possible notations health care professionals might use. In particular, the use of non-standard abbreviations in the free text fields of the medication schedule could result in complexity factors not being detected in the automated analysis. Third, the tool was tested with a limited number of exemplary medication schedules. It is, therefore, possible that not all possible cases are considered when specifying criteria for the automated detection of individual factors; hence, some complexity factors may have escaped detection by the electronic tool. However, the tool will be tested in a pilot study with a large number of actual medication schedules issued by different prescribers allowing further optimisation of the electronic tool and adaptation to routine care.

We developed an algorithm-based electronic tool that covered a wide range of different factors that are known to increase complexity of drug treatment. The majority of factors could be identified in an automated analysis by the tool and key questions could be phrased to assess whether these complexity factors indeed pose a problem for a specific patient. Relevant complexity factors that were not suitable for an automated detection were considered by distinct questions and a general, open-ended question. The electronic tool combines an automated screening with a personalized intervention to comprehensively assess, mitigate, or antagonize treatment complexity. This electronic tool is designed to tailor medication regimens to the needs of the individual patient and is now ready for testing in a prospective pilot study.