FormalPara Key Points

Older patients with polypharmacy are exposed to a risk for pharmacodynamic drug interactions.

Prescribers seem to respond to warnings for QT prolongation given by a clinical decision support system.

1 Introduction

The number of patients with polypharmacy is rising [1]. About 10.7% of all Swedish individuals aged older than 75 years take ten or more medications every day. This entails an increased risk of adverse effects and a high number of hospitalisations. Patients who take six to ten medications every day are 48% more likely to experience adverse effects compared with patients who take five or fewer medicines [2]. The number of medicines used by patients increases the risk for potentially inappropriate prescribing and potential prescribing omissions. This is especially a challenge for older patients who are also more prone to experience adverse effects [3, 4]. Of note, in many patient populations, polypharmacy is a necessity and patients benefit, for example multimorbid patients or patients with conditions typically requiring treatment with several substances such as hypertension and or diabetes mellitus [5]. However, polypharmacy can also have negative effects such as increased healthcare utilisation, cognitive impairment, fall injuries and other adverse effects, and increased mortality [6]. Many of these adverse effects may be the result of pharmacodynamic drug–drug interactions, for example anticholinergic effects that are especially common in older patients. Many medications bear the risk of prolonging the QTc interval. Other risk factors for QT prolongation are old age, QT prolongation at electrocardiogram (ECG) baseline, female sex, electrolyte disturbances (hypokalemia, hypocalcemia and hypomagnesemia), bradycardia and hereditary cardiac diseases. Prolongation of the QTc interval increases the risk for torsades de pointes (TdP), leading to ventricular tachycardia and fibrillation and sudden cardiac death [7]. In Sweden, 410 TdP cases associated with the use of medicines with a known risk for TdP were registered between 2006 and 2017 [8]. However, this study likely underestimated the real number of TdP cases because to diagnose TdP the recording of this dangerous arrhythmia is needed.

Increased use of medications will increase the risk for adverse effects and there is a need for tools and guidance on how to handle this including prescribing guidelines, medication reviews and automated decision support systems that recognise risk factors and warn prescribers [9]. The Janusmed Risk Profile was introduced in 2018 and aimed to support clinicians with a special focus on adverse effects caused by pharmacodynamic interactions.

2 Objectives

The main objective of this study was to investigate to what extent older persons (aged 65 years or older) with polypharmacy (defined as using five or more medicines) are exposed to risks for nine adverse drug effects included in the Janusmed Risk Profile. A secondary objective was to investigate any changes in the risk scores after implementation of the clinical decision support system (CDSS) Janusmed Risk Profile in the main electronic health record (EHR) system in Stockholm.

3 Methods

3.1 Design, Data Sources and Study Population

We conducted a cross-sectional retrospective register study based on data from the Stockholm Region administrative database (VAL). The database includes all prescription medicines purchased at the pharmacy by patients living in the Stockholm Region and information about the healthcare provider who prescribed the medication. The database covers drugs prescribed in primary, secondary and tertiary care but does not cover drugs used when the patient is in hospital care, or over-the-counter drugs. Patients using five or more medications, defined as unique active substances, during the study periods were included in the study. Each dispensed medication between 2011 and 2018, prescribed by healthcare providers using the main EHR system in Stockholm, TakeCare (in primary, secondary and tertiary care), to patients aged 65 years or older were retrieved from the VAL database.

3.2 Study Periods

We used cross-sectional data from period 1 (November 2016–February 2017) and period 2 (November 2017–February 2018) to calculate the percentage of patients with combinations classified as increased risk (moderate or high, and high) and any changes thereof between the two. All unique active substances purchased in these 4-month periods were included in the analysis. To illustrate longitudinal changes in the risk score pattern, we show the trend for each of the categories from 2011 to 2018 using the same 4-month periods (November–February) as in the analysis between period 1 and period 2.

3.3 Janusmed Risk Profile

The Janusmed Risk Profile is a CDSS designed for the pharmacological risk assessment of a patient’s complete medication list. The system provides a risk profile based on nine common or serious adverse events relevant in older patients. The development and the selection of ADRs included have been described earlier [10]. The target users of the CDSS are healthcare professionals performing medication reviews, prescribing new medications or evaluating whether a patient’s clinical symptoms may have been caused by the prescribed medication. The Janusmed Risk Profile warns for potential risks when two or more medicines enhance each other’s effects in concomitant treatment. It contains nine adverse-effect risk categories: anticholinergic effects, constipation, sedation, orthostatism, haemorrhage, serotonergic effects, risk of seizures, QT prolongation and renal toxicity. The Janusmed Risk profile covers almost all approved substances at the Swedish market. The assessment process of the CDSS takes active substances and formulations into account. Each substance is evaluated for each risk category and has been assigned a risk value between 0 and 3. For renal toxicity, only a value of 0 or 1 has been assigned because a substance either has this characteristic or not. Predefined algorithms calculate the total risk score for each category using the risk values for each substance used by the patient. As a result, the interface presents the risk level for each category. The risk levels are also classified between 0 and 3, which corresponds to no risk, low risk, moderate risk and high risk, and marked with pertinent colours for a user-friendly presentation at the web interface, see Fig. 1. First, the user gets a schematic overview of all risks and medications/substances and their contribution to each specific risk category (see Fig. 2).

Fig. 1
figure 1

Risk levels according to the Janusmed Risk Profile indicated by colour and italic number

Full size image
Fig. 2
figure 2

Schematic overview of the risks for a patient using citalopram, morphine, hydroxyzine, St Johns wort, apixaban, enalapril and omeprazole

Full size image

In addition, the interface presents information about each risk category that has triggered the warning signals, including medical consequences and a short recommendation/advice on the subject (Fig. 3). The Janusmed Risk Profile does not include warnings for pharmacokinetic drug–drug interactions but is used alongside Janusmed Interactions that covers all pharmacokinetic drug–drug interactions.

Fig. 3
figure 3

Example of risk text for haemorrhage for a patient using citalopram, St. Johns wort and apixaban. A summarised value of 5 gives a high risk for haemorrhage. INR international normalised ratio, SNRI serotonin and norepinephrine reuptake inhibitor, SSRI selective serotonin reuptake inhibitor

Full size image

The Janusmed Risk Profile is implemented in the EHR system used in this study (TakeCare). It is part of a toolbar alongside other CDSSs such as pharmacokinetic interactions and drug dosing in renal insufficiency. The Risk Profile button is coloured according to the highest individual risk level in any of the nine categories and may change its colour as soon as a new medication is prescribed if it affects the risk levels.

3.4 Risk Profile Calculation

The patient’s medication list for the 4-month period was reduced to the unique substance level. Topically administered medicines were excluded from analysis as they were, at the time of the data analysis, not classified in the Janusmed Risk Profile. The patient’s risk assessment level (none, slight, moderate, high) for each of the nine properties was calculated according to the algorithm used in the Janusmed Risk Profile. This was done separately for the two study periods. The risk for seizures is only classified as slight or high and therefore is only included in the analysis of high risks. As the risk profile is continuously updated with no old versions available, we used the current version (downloaded 2022-09-07).

3.5 Statistics

Patient characteristics are described as median and interquartile range. The proportion of patients with a risk classified as high or moderate and only high was calculated and is presented as percentages. Differences in the proportion of patients with a high and moderate risk and only a high risk between periods 1 and 2 were analysed using a two-sided Fisher’s exact test with a two-sided p < 0.05 considered as statistically significant. A Bonferroni–Holm correction was performed to adjust for multiple testing. Data analyses were performed using R version 4.0.2 [11].

4 Results

In total, 127,719 patients in period 1, and 131,458 patients in period 2 met the inclusion criteria. Patient characteristics are summarised in Table 1. Around one third were exposed to a moderate to high risk for haemorrhage, constipation and orthostatism. Between 13% and 20% of patients had a high or moderate risk for sedation and QT prolongation. Serotonin syndrome and renal toxicity were uncommon, with less than 2.5% of patients being exposed. When analysing the proportion of patients with a high-risk level, the pattern was similar with constipation being the most common risk (19%) followed by haemorrhage (13%). Proportions are presented in Table 2.

Table 1 Patient characteristics
Full size table
Table 2 Proportion of patients at moderate or high risk and high risk and their changes between period 1 and period 2
Full size table

There were statistically significant changes between periods 1 and 2 in the proportion of patients with a moderate-to-high risk for the categories haemorrhage, QT prolongation and sedation. The risk increased for haemorrhage (from 34.7 to 35.5%) and decreased for QT prolongation (from 17.2 to 10.8%), and sedation (from 15.4 to 14.9%). When analysing patients with a high risk only, the change was significant for haemorrhage (from 13.4 to 13.1%), QT prolongation (from 4.6 to 2.3%) and sedation (from 10.3 to 9.9%) with a lower risk for each of these categories in period 2. With regard to orthostatism, we observed a numerical decrease in the proportion of patients with a high risk, but the results missed the critical significance level after the Bonferroni–Holm correction by a small margin. There were no significant changes for all the other risk categories (see Table 2). The Electronic Supplementary Material (ESM) lists the most common drugs that contributed to the different risk categories. The proportion of patients with a high risk for each property from 2011 to 2018 is illustrated in Fig. 4. There were no marked changes over time.

Fig. 4
figure 4

Trends in the proportion of patients at high risk for each property included in the Janusmed Risk Profile. No. 6 on the x-axis is period 1 in the study (November 2016–February 2017) and no. 7 is period 2 (November 2017–February 2018). The values 1–5 show the proportions observed for the 5 years prior to the study periods (e.g. 1 is November 2011–February 2012)

Full size image

5 Discussion

This study is the first to report the risk scores according to the Janusmed Risk Profile among older patients. To the best of our knowledge, there is no comparable comprehensive analysis software of pharmacodynamic risks associated with polypharmacy. We show that more than one third of patients using five or more medications had a high or moderate risk for haemorrhage, constipation or orthostatism. A high-to-moderate risk for sedation and QT prolongation was also common. Thus, this study shows that pharmacodynamic interactions are common among older patients with polypharmacy.

5.1 QT Prolongation

The most notable difference between the time periods was seen for the risk of QT prolongation where the proportion of patients with a high or moderate risk decreased from 17.2% in period 1 to 10.8% in period 2, after the introduction of the CDSS. Accordingly, the proportion of patients with a high risk decreased from 4.6% to 3.3%. In contrast to several other factors that can be easily monitored and are checked at routine visits, ECG controls are not regularly performed if there is no suspicion of cardiac or cardiovascular disease. For the clinician, it may also be difficult to gather helpful information regarding drug effects on the QT interval. The Summary of Product Characteristics, easily accessed by the open Physicians’ Desk Reference (FASS) in Sweden, can include information and warnings, but the text may be confusing and difficult to interpret, being for example too generalising with unclear recommendations. Other existing international web sources such as Credible Meds [12] are probably not commonly known and require additional work by the prescriber. Thus, a CDSS integrated into the EHR system would be expected to greatly facilitate the access to critically evaluated information and increase the prescribers’ awareness of the risks of QT prolongation.

Another possible reason for the observed change in the risk for QT prolongation is that clinicians reacted to previously unknown high-risk combinations, for instance by stopping drugs or switching to alternative substances not bearing that possible risk. Clinicians may be especially prone to solve the risk for QT prolongation because of its potentially detrimental consequences. This is supported by the unpublished results from a survey we sent out to users of the Janusmed Risk Profile. Physicians responded that QT prolongation was the risk category where they were most prone to change the medication (see Appendix 2 of the ESM). As QT prolongation is not always a class effect, but quite often substance specific, alternatives that are less likely to prolong the QT time are often available.

While the results from this study suggest that the introduction of the Janusmed Risk Profile CDSS may have contributed to lowering the potential risk of QTc prolongation and hence arrhythmias in the older population, it is impossible to exclude other contributing factors based on limitations inherent to the retrospective study design. There have been repeated information campaigns about drugs and the risk for TdP from national agencies and drug and therapeutic committees. Some illustrative examples include hydroxyzine and (es)citalopram, which were commonly prescribed among older patients and generally considered safe until pharmacovigilance studies highlighted the risk of QT prolongation [13, 14]. These warnings were issued a few years before this study, but a time delay with regard to their impact on clinical practice cannot be fully excluded. One could hypothesise that the decreased concomitant prescription of drugs with the potential to cause QT prolongation could be due to a combination of increased knowledge and awareness of the risks and an upbuilt need for an easily available information source, i.e. that the Janusmed Risk Profile was launched at an optimal time.

5.2 Other Risks Before and After Integration of the Janusmed Risk Profile

We observed somewhat inconclusive results regarding haemorrhage: a minor decrease in high risk, but a minor increase for high or moderate risk. In the algorithm, substances with the highest individual risk, including anticoagulants, are assigned a moderate risk for haemorrhage. The addition of another drug with any risk level for haemorrhage will increase the sum risk level to high. One of the major changes in the recommendations in Stockholm from 2016 to 2017 [15] was that warfarin was replaced by apixaban as a first-line anticoagulant. One possible explanation for the increase in moderate risk between period 1 and period 2 is the introduction of and official recommendations to use direct oral anticoagulants over warfarin because many patients were deemed unsuitable for therapy with warfarin, but not for treatment with direct oral anticoagulants [15]. The total number of prescriptions for warfarin/direct oral anticoagulants per patients increased from period 1 to 2 (see Appendix 1 of the ESM). Of note, a switch from warfarin to apixaban within either of the study periods would, however, result in a high-risk warning and thus not contribute to the increase in the moderate-risk group. This is an important limitation of this study.

The proportion of patients with a high risk for sedation was decreased from 10.3% in period 1 to 9.9% in period 2. As the use of sedating medications is associated with the risk for falls among old people [16], a decrease in the risk would potentially be of clinical relevance. The observed change although statistically significant was rather small and with uncertain clinical relevance. This decrease could also be due to other factors such as recommendations to reduce the prescription of sedating medications in older patients [17]. There was no significant change in the proportion of patients with a high or moderate risk or a high risk alone for anticholinergic effects, constipation, renal toxicity and serotonin syndrome.

As the Janusmed Risk Profile calculates the sum of the risks of individual substances, the risk score will only decrease if one or more of the medicines are stopped or replaced with another medicine with a lower risk for the specific risk category. Other possible measures taken by the physician such as dose adjustment cannot be detected in the study, neither would a closer follow-up (i.e. checking laboratory values, ECG) or prescription of medications to treat the adverse effects such as constipation be identified.

The Swedish National Board of Health and Welfare has for many years published recommendations of which medicines should be avoided in older people. Furthermore, in patients aged 75 years or older using five or more medications, a simple medication review should be performed [17]. The latest guidelines on that matter were released in 2017, which may also have impacted on prescribing patterns. These guidelines regularly contain checklists for common adverse effects among older patients including sedation, constipation, orthostatism and some typical anticholinergic effects. In that context, the Janusmed Risk Profile CDSS could be of great help to identify those risks and be a much quicker alternative to searching other sources such as the Summary of Product Characteristics or guidelines. In contrast to most other CDSS, the Janusmed Risk Profile does not only analyse drug pairs, but provides an analysis of the whole medication of an individual patient. Its algorithm combines the risks of all medications for a certain risk category and calculates a sum score, which provides much more intuitive information for clinicians.

5.3 Study Strengths and Limitations

The study design has both strengths and limitations. Retrospective use of registers makes it possible to include large patient cohorts, as in this case more than 120,000 patients. There is no risk for the design to interfere with the outcome. We do not have any missing data in the prescription database; however, we lack information about medicines used over the counter and prescribed by healthcare personnel not using the EHR system TakeCare. We also do not have diagnostic data. The retrospective design precludes randomisation and the use of a contemporaneous control group. Hence, controlling for confounding factors is mostly impossible. A critical issue of the methodology is that we do not know what medications have actually been taken concomitantly. As explained in detail above, a switch of an anticoagulant during the 4-month study period would result in a high-risk warning that we cannot differentiate between concomitant and subsequent intake. This is also true when patients switch to other treatments such as antidepressants within the study period. Another limitation is that we cannot exclude the influence of general warnings issued for certain drugs during the study. The Janusmed Risk Profile does not take doses or monitoring of therapy (e.g. regular ECGs) into account. A 4-month period was chosen to cover medicines used for chronic diseases because patients usually purchase medicines for a 3-month period at a time because of the Swedish reimbursement system. Hence, there may be an overestimation of the risks. An additional limitation, which may underestimate the true risk, is that we only included prescriptions from healthcare units using TakeCare. This is the main EHR system in the Region Stockholm, but a minority of healthcare units use other systems. However, the Janusmed Risk Profile was only implemented in TakeCare. Thus, in some patients, who were also treated in those units, we may have missed some medications and we may have underestimated certain risks. In that context, we have previously shown that many drug–drug interactions are due to prescriptions from different prescribers [18]. The Risk Profile version used in this study is newer than the one included in TakeCare before period 2 and some minor changes have been made since then. This may influence risk-level calculations and the clinical decisions of users in some cases.

6 Conclusions

Pharmacodynamic interactions are frequent in older patients with polypharmacy. Whereas most CDSS only analyse drug pairs, the Janusmed Risk Profile is the first to provide a comprehensive analysis of all medicines prescribed to a patient. Introduction of the software led to a pronounced reduction in the combination of medicines associated with the risk for QT prolongation, while it only affected other risk categories marginally.