A Complex Intervention to Prevent Medication-Related Hospital Admissions

Abstract

The collaborators of this article are listed in the eBox.

Background

Children are often treated off-label and are at a disadvantage in pharmacotherapy. The aim of this study was to implement and evaluate a quality assurance measure (PaedPharm) for pediatric pharmacotherapy whose purpose is to reduce medication-related hospitalizations among children and adolescents.

Methods

PaedPharm consisted of the digital pediatric drug information system PaedAMIS, pediatric pharmaceutical quality circles (PaedZirk), and an adverse drug event (ADE) reporting system (PaedReport). The intervention was implemented in a cluster-randomized trial (DRKS 00013924) in 12 regions, with a pediatric and adolescent medicine clinic in each and a total of 152 surrounding private practitioners, in 6 sequences over 8 quarters. In addition to the proportion of ADE-related hospital admissions (primary endpoint), comprehensive process evaluation included other endpoints such as coverage, user acceptance, and relevance to practice.

Results

41 829 inpatient admissions were recorded, of which 5101 were patients of physicians who participated in our study. 4.1% of admissions were ADE-related under control conditions, and 3.1% under intervention conditions (95% CI: [2.3; 5.9] and [1.8; 4.5], respectively). A model-based comparison yielded an intervention effect of 0.73 (population-based odds ratio; [0.39; 1.37]; p = 0.33). PaedAMIS achieved moderate user acceptance and PaedZirk achieved high user acceptance.

Conclusion

The introduction of PaedPharm was associated with a decrease in medication-related hospitalizations that did not reach statistical significance. The process evaluation revealed broad acceptance of the intervention in outpatient pediatrics and adolescent medicine.

When it comes to drug therapy, children are at a disadvantage in terms of safety and proof of efficacy. Between 42 and 90 percent of children and adolescents are treated with drugs off-label in the inpatient setting and between 46 and 64 percent in the outpatient setting (1– 6). Reasons for this include a lack of clinical studies due to methodical and ethical challenges in the pediatric subpopulation and the limited economic interest of pharmaceutical companies (7– 9). Furthermore, the off-label use of drugs is associated with an increased risk of adverse drug events (ADEs) (10– 14).

About three to five percent of all hospitalizations involving children are the direct result of an ADE (11, 12, 15). A study from the UK shows that 22.1% of these were avoidable and should therefore be regarded as medication errors (MEs) (11).

Children are excluded from the monitoring system for approved medications when it comes to off-label use. The continuous spontaneous reporting of ADEs by attending doctors is therefore an important instrument to generate drug safety signals, especially in pediatrics.

Despite the increase in drugs approved specifically for children as a result of the requirements of the EU Pediatric Regulation, off-label use remains a challenge for pediatric pharmacotherapy, now and in the future. Not every off-label use is associated with the same risks (16, 17). That is why the ready availability of existing evidence and knowledge of the special features of pediatric pharmacotherapy are particularly important.

The aim of the project “KiDSafe—Improving drug therapy in children and adolescents by increasing medication safety” was to lower the proportion of ADE-related hospitalizations by introducing a structured quality assurance measure (PaedPharm), comprising a pediatric drug information system (PaedAMIS), amongst other things, and to improve pediatric medication safety.

The present article summarizes the key aspects of the study. Additional available information and results may be found in the publicly accessible outcome report (18).

Methods

PaedPharm

PaedPharm is a model of care comprising three modules (PaedAMIS, PaedZirk, and PaedReport) to provide and train healthcare professionals with evidence-based information about drug therapy in children and adolescents (box).

BOX. The components of the intervention PaedPharm.

• Module 1 PaedAMIS*

Digital pediatric drug information system

The module PaedAMIS is a web-based platform with comprehensive information on drug therapy in children. The database comprises individual active ingredient monographs, each containing dosage recommendations, including their regulatory approval, available preparations appropriate for children, and other instructions for use (warnings, contraindications, interactions). The dosing instructions were generated by a standardized, systematic approach and compiled according to an expert-based consensus procedure. They are based on the best available evidence, with the underlying literature provided where regulatory approval was absent. Furthermore, dosage information is organized according to indication and route of administration/dosage form. Apart from individual active ingredient monographs, the database also provides further information on pediatric issues, for example, dosage forms suitable for children and reference percentiles. Only the participating physicians had access to the database during the study. Login details were not made available until immediately before the training phase. Use of the database was continuously available throughout the entire project duration.

• Module 2 PaedZirk

Pediatric pharmaceutical quality circles

Regional clinical pediatric pharmaceutical quality circles for physicians in private practice. Topics: use of drugs for children (pediatric pharmacotherapy, prescriptions of psychopharmaceuticals, and much more), training material (for example, regular information letters on selected topics), discussion of selected pharmacological and pharmaceutical issues in pediatric pharmacotherapy, the medication process, possible sources of errors, and recognition of adverse drug events when prescribing medicines.

• Modul 3e PaedReport

A national adverse drug event reporting system for off-label prescriptions in children

Training of physicians to recognize and report adverse drug events, including medication errors in outpatient settings, to the Drug Commission of the German Medical Association.

*PaedAMIS is the result of Measure 16 of the Action Plan for Medication Safety (MS) 2013–2015 of the Federal Ministry of Health (BMG) to improve medication safety in Germany and has been freely available at www.kinderformularium.de since 01/2021

Study design and implementation

PaedPharm was developed and evaluated during the present study (DRKS—German Clinical Trials Register DRKS 00013924) in accordance with current international recommendations of the Medical Research Council UK for complex interventions (19, 20). Approvals were obtained from the responsible ethics committees prior to the start of the study.

The present publication is oriented on the CONSORT guideline for stepped wedge cluster randomized trials (cRCT; [21]). The intervention was implemented throughout Germany in 12 regions (clusters) together with specialist doctors in private practice for pediatrics and adolescent medicine, child psychiatry, and general medicine, in a stepped wedge cluster randomized trial using a cross-sectional design. Each cluster consists of a pediatric and adolescent medical clinic (university hospital or municipal hospital) and their regionally referring participating specialists in private practice.

PaedPharm was introduced in a stepwise manner in six different sequences (figure 1). The twelve clusters were randomly allocated to the six sequences (eMethods section). After a three-month training phase (T), two clusters each changed on a quarterly basis from the control phase (C) to the intervention phase (I). Thus, clusters which had not yet been introduced to the intervention served as control regions for those clusters where introduction of the intervention had already taken place. At the start of the intervention phase, all participating doctors received exclusive access to PaedAMIS via a username and a password and were invited to regular quality circles.

Figure 1.

Study design – time-based allocation of the various intervention phases in the sequences

I, intervention phase (intervention effect 1); C, control phase; T, training phase (intervention effect 0,5)

Evaluation methods, endpoints, and sample size

Data to assess the benefits of PaedPharm were collected in the participating clinics. The primary endpoint was the proportion (%) of ADE-related hospitalizations in all non-elective hospital admissions where assignment to participants was possible. The intention was to reduce this proportion by one third (from 3% to 2%) (12). Calculated for 12 clusters, six sequences, a duration of seven quarters, and a power of at least 80%, it was planned to recruit 240 doctors (eMethods). As part of a quantitative and qualitative process evaluation, a secondary aim was to observe the implementation process and direct intervention aspects, for example, user acceptance.

Data acquisition and instruments for assessment

Between 01 July, 2018, and 30 June, 2020, each non-elective admission to the participating clinics which fulfilled the inclusion criteria was identified (table 1), and age on admission, sex, medication history, and diagnoses (according to ICD-10) were documented.

Table 1. Inclusion and exclusion criteria at clinic and case levels.

	Clinic	Case
Inclusion criteria	–  IT infrastructure available – Previous experience with the electronic acquisition of patient data – Geographic location (urban/rural)	Cases of patients admitted non-electively as inpatients who were below the age of 18 years at admission; precondition: use of a drug a) in the previous seven days prior to admission or b)  at least five days in the previous eight weeks prior to admission or c) taken/used for long-term therapy
Exclusion criteria		Oncological patients admitted as inpatients for a suspected adverse drug event and who had been under inpatient oncological treatment during the previous four weeks

On establishing a positive medication history, a check was made for the presence of an adverse drug event (ADE). For this purpose, an algorithm designed specifically for the study to detect cases of ADE was used (22). Once an ADE was suspected and after obtaining consent from the parents, the case was transferred in pseudonymized form to the ADE coordination unit (ADE-C). The region of origin and, thus also, the intervention status were rendered unrecognizable to the coordination unit.

Suspected cases of an ADE without parental consent were forwarded to the Drug Commission of the German Medical Association (AkdÄ) after an initial causality check by the local study team (presence of at least a temporal relationship according to the WHO-UMC system [23]). Suspected ADEs were examined by both assessor authorities (ADE-C and AkdÄ) using established algorithms and definitively assessed (ADE suspicion “confirmed”/“not confirmed”; [23]). Training and regular exchange between the two assessor authorities ensured that evaluation of suspected ADEs was comparable between both authorities. The result was recorded on a local level (figure 2). Hospital admissions due to medication errors (ME) were always rated as drug related. Cases were communicated to the evaluating institution on a monthly basis in pseudonymized, anonymized or aggregated form, depending on whether or not consent had been given.

Figure 2.

Flow diagram of data acquisition process conducted in the clinics

* Suspected medication-related hospitalization

AkdÄ, Drug Commission of the German Medical Association; ADE, adverse drug events; ADE-C, ADE coordination unit

Statistical analysis

For the primary endpoint, the proportion of ADE-associated admissions in all hospitalizations by the referring and participating private practitioners under control and intervention conditions was compared. A logistic model for correlated data was fitted to these data, taking into account the cluster-randomized stepped-wedge design (eMethods). The intervention effect is reported as an odds ratio with a 95% confidence interval. Missing data were replaced using a multiple imputation procedure (eMethods).

Process evaluation

Questionnaires asking about satisfaction with the quality circles (PaedZirk) and the number of spontaneous reports from participating practices (PaedReport) were used to describe the quality of implementation. At the end of the evaluation period, a standardized questionnaire was sent to all participating physicians asking about acceptance, satisfaction, relevance to practice, and sustainability of PaedPharm (efigure 1).

eFigure 1.

Results

The CONSORT flow diagram (efigure 2) provides an overview of inclusion and continuation of all participating physicians and of all hospitalizations.

eFigure 2.

Participating physicians

Out of 1386 physicians contacted, 160 (240 were planned) were prepared to participate in the trial (136 from pediatric and adolescent medicine, ten general medicine, 14 pediatric and adolescent psychiatry and psychotherapy). Eight doctors stopped participating even before introduction of the intervention, allowing cases from 152 doctors to be included (efigure 3).

eFigure 3.

Hospital admissions

Altogether, 41 829 patient cases were included (efigure 3), of which 5101 admissions were assigned to participating physicians. It was not possible to compile a participation status in 209 cases (control phase n = 106, training and/or intervention phase n = 103). In 55 cases, a definitive ADE assessment by the ADE assessor authorities (control phase n = 40, training and/or intervention phase n = 15) was not possible. These missing data (n = 264) were imputed for the regression analysis. Table 2 shows the characteristics of this identified patient population.

Primary endpoint

Altogether 77/2497, 36/775, and 66/1829 ADE-related hospitalizations from participating physicians in control, training, and other intervention phases were identified, respectively. Thus, according to period-corrected estimates of the population-based mean and under control conditions, 4.1 % of admissions were ADE-related. This figure was 3.1% (95% confidence interval: [2.3; 5.9] and [1.8; 4.5], respectively) under intervention conditions. The model-based comparison resulted in a population-based odds ratio (= intervention effect) of 0.73 ([0.39; 1.37]; p value 0.33). A second post-hoc analysis of the primary endpoint was conducted to compensate for the too low number of participating physicians and to use the collected data more efficiently. All recorded non-elective hospitalizations were taken from control quarters and used for this analysis, which therefore also included those from non-participating doctors. This allowed 57% of all recorded admissions to be included in this analysis. After these adjustments and in the population-based mean, 3.7% of admissions were ADE related under control conditions, and 3.0% under intervention conditions. The model-based comparison resulted in a population-based odds ratio comparable to the main analysis (= intervention effect) of 0.80 ([0.52; 1.24]; p value 0.31).

Process evaluation—Pediatric drug information system—PaedAMIS

Altogether, 96 physicians participated in the final survey. Of these, 70% stated they were “partially” (n = 22/92) or “completely” (n = 42/92) satisfied with PaedAMIS (on a five-point Likert scale). Regarding relevance to practice, 40% (“fully or partially relevant”, n = 12 + 25/92) stated that PaedAMIS was a great help in daily work, while 63% of respondents (n = 57/91) said they consulted PaedAMIS less than once a week. Three participants (3/91) used PaedAMIS several times a day. Thirty-three of 81 (41%) respondents had subsequently altered their prescriptions in five to 10 percent of cases. For the majority of respondents, the information in PaedAMIS had resulted in at least one alteration in their prescribing pattern (n = 66/81).

Process evaluation—Quality circles—PaedZirk

During the whole project period, 45 of 72 scheduled quality circles were held in presence and, due to the pandemic, one as a live webinar. The mean number of participants was 7.3 person/invitation. Twenty-five scheduled quality circles had to be canceled because of the corona virus pandemic of 2020. The webinar invitation was offered for the final sequence because this one was particularly affected by contact restrictions (efigure 4). The quality circles were considered relevant to practice by 81% (78/96) of the participants. Participation in the seminars changed their confidence in drug prescription for 67% (64/96). The quality circles were rated as “good” to “very good” in every category. On average, the participants also gave a grade 1.3 to the category “relevance to practice”.

eFigure 4.

Process evaluation—ADE reports—PaedReport

A total of 28 spontaneous reports were recorded during the evaluation period. With altogether 160 participating physicians, that was an annual reporting rate of 0.088 spontaneous reports per physician.

Process evaluation—Sustainability

Of 96 respondents, 72 would still use PaedAMIS and 73 would attend the quality circles after the end of the project. Half of them (n = 48) reported that something had changed sustainably with regard to processes of prescribing medications. Forty-five percent (n = 48) said that structures and/or processes pertaining to reporting ADEs had changed. Sixty percent (n = 58) said that PaedPharm had achieved a sustainable change in their awareness for ADEs. Sixty-six percent (n = 63) intended to report suspected cases of ADE more often in the future. Eleven percent (n = 11) of the respondents reported no change.

Discussion

The present study examined for the first time in a large patient population in Germany the benefit of a multimodal intervention for pediatric medication safety. A little more than four percent of all non-elective hospitalizations were due to ADEs, confirming the results of international studies (11). With regard to the benefit of PaedPharm, there was a relative risk reduction of -26%, which was close to the assumption of -33% made during sample size calculation.

The lack of statistical significance is primarily due to the small number of participating private practitioners (240 planned versus 152 effective participants). Only 11.5% of the approached physicians were prepared to participate in the healthcare study. The actual participation rate was similarly low to that of other studies (24). At 13%, the proportion of admissions from participants lagged behind the planning assumption of 40%.

Altogether 25 of 72 scheduled quality circles (35%) had to be canceled because of the coronavirus pandemic and the associated lockdown measures. The quality circles were an important component of the intervention, and this was possibly a reason for the somewhat weaker observed intervention effect (-26 %) in comparison with what had been assumed (-33 %). However, sensitivity analyses showed that the data of the last period contributed little to the overall outcome.

The process evaluation provided information indicating acceptance and benefit of the intervention. More than two thirds of participants confirmed that they intended to continue using Paed AMIS in the future. Paed AMIS was still in the development phase when the KiDSafe study was started, and only around 100 monographs were available. This number had already risen to 401 by the end of the project (June 2021). This included monographs which had been prioritized following discussions with the physicians in the quality circles. In the meantime, the pediatric drug information system is freely available to all healthcare professions under the name Kinderformularium.DE (www.kinderformularium.de) and is constantly being expanded and updated.

The available quality circles (PaedZirk) were regarded as extremely relevant to practice and were well attended. The personal participation of experts in the field of pediatric and adolescent pharmacotherapy was particularly appreciated. The results reveal the considerable importance of expertise in developmental pharmacology for appropriate pharmacotherapy. Regular further education courses together with appropriate expertise could therefore continue to play an important role in the future.

The spontaneous reporting of ADEs is of particular importance in pediatrics, given the common off-label medication use and associated lack of systematic monitoring following marketing authorization. It must be assumed that the high degree of off-label use together with its associated insecurity amongst doctors and anticipated litigation/legal consequences in the event of ADEs have a negative impact on the reporting rate. In Germany in 2008, the reporting rate for children was only 45% of that for adults (reports per million population; [25]). According to the annual report of the AkdÄ, 231 spontaneous reports for the age group children and adolescents up to the age of 18 years were registered in 2018 (26). During the same year, there were 5776 specialists for pediatric and adolescent medicine practising in Germany and 996 specialists for pediatric and adolescent psychiatry (27). Based on these figures, this results in an estimated annual reporting rate of 0.018 notifications per private specialist. The reporting rate during the KiDSafe study was 0.088, indicating an almost fivefold increase in the rate. That could demonstrate that PaedPharm succeeded in creating an increased awareness and attention for ADEs.

Strengths and weaknesses

It can be assumed that the study design and the selected primary endpoint resulted in a high internal validity of the results. The selected design allowed for adjustments for period effects. Thus, the anticipated increase in cases of suspected ADE once data acquisition had started could also be taken into consideration.

The identification of potential ADE-related hospitalizations was standardized by a clinical algorithm, and validation of suspected cases was achieved by independent, blinded experts (UAE-C) and the AkdÄ. The strict separation between implementation in private practices and the fully decoupled evaluation in clinics for pediatric and adolescent medicine also promised a high degree of validity, as did the institutional distinction between implementation and evaluation.

The lower-than-anticipated participation rate of physicians in private practice and the low consent rate for a high-quality assessment of suspected ADEs in a reference center were possibly obstacles for an empirical proof of efficacy of the intervention. Another factor was that restrictions imposed by the coronavirus pandemic prevented implementation of the project to the planned extent.

Summary and conclusions

With PaedPharm, a multimodal intervention to improve medication safety in children and adolescents was implemented and evaluated for the first time in Germany. The intervention resulted in a statistically not significant reduction of drug-related hospitalizations.

The process evaluation revealed the potential for PaedPharm to raise awareness for the correct use of drugs in children and adolescents and to improve the reporting rate of ADEs.

Supplementary Material

eMethods

Statistical methods

Randomization

The twelve study regions were grouped into associated pediatric clinics of six larger and six smaller institutions each, based on the number of inpatient admissions in the previous year, and then allocated to six sequence groups, each with one small and one large institution, using computer-generated random numbers. Allocation only took place after all twelve centers had formally consented and been included in the study.

Sample size calculation

For the purposes of sample size calculation, it was assumed that the risk of an ADE varies between the clusters by a factor of between 0.80 and 1.22, with a 95% probability. This results in an intracluster correlation coefficient (ICC) of 0.0003. It was also assumed that 7500 non-elective admissions could be registered per quarter of the year, of which another 40% (i.e., 3000 per quarter) could be allocated to participating physicians. Under the assumption that 3%, 2.5%, and 2% of admissions are ADE-associated in control, training, and intervention quarters, respectively, then 315, 75, and 150 ADE-associated admissions, respectively, would be expected. The assumed intervention effect can be determined with a power of at least 80%. Participation of 240 doctors was planned to achieve these case numbers. We used the R package swCRT design for sample size calculation (e1).

Details of the statistical method for data analysis

A generalized linear model for correlated longitudinal data with logistic link function was fitted for the primary endpoint, based on the used stepped wedge design. The model contains a fixed period effect and therefore allows a time-trend adjusted estimate of the intervention effect. The model was fitted using generalized estimating equations (GEE). An exchangeable correlation structure at cluster level was assumed. The odds ratio calculated in this way is to be interpreted as a population-based, or marginal, odds ratio. It may be interpreted as a quotient between the incidence rate (or relative risk) of ADEs in the intervention setting and the incidence rate in the control setting. The standard error and resulting confidence interval (CI) were calculated using the Mancl and DeRouen approach (e2). This procedure is also robust to the fact that a small amount of inflation was to be expected from multiple hospitalizations of the same child.

Handling missing data

In cases where there was no conclusive evaluation of whether an ADE was present (control phase/intervention or training phase n = 40/15) or for which it was not possible to allocate with certainty whether children admitted as inpatients were treated by participating physicians (control phase/intervention or training phase n = 106/103), missing values were imputed multiple times (drawn at random several times) to prevent these missing values from distorting the estimate. The distribution of the existing assessments was used for the imputation of the ADE assessments, stratified according to the type of assessment pathway (via ADE-C or AkdÄ) and the intervention status of the clinic. The clinic and the intervention status of the clinic were used to impute the participation status (of the associated physician). The analysis model was adapted for each of a total of 25 randomly imputed data sets. The resulting intervention effect estimator (log odds ratio) and its standard error were combined using the Little and Rubin method (e3).

Table 2. Characteristics of the patient population (data from 10 of 12 clinics*).

	Control		Intervention		Total
Age group (n = 34 501)
0 to 3 years	7644	43%	7236	44%	14 880
3 to 6 years	2522	14%	2522	15%	5044
6 to 9 years	1537	9%	1389	8%	2926
9 to 12 years	1585	9%	1414	9%	2999
12 to 15 years	2047	11%	1708	10%	3755
15 to 18 years	2553	14%	2344	14%	4897
Total	17 888	100%	16 613	100%	34 501
Sex (n = 34 467)
Male	9066	51%	8424	51%	17 490
Female	8814	49%	8163	49%	16 977
Total	17 880	100%	16 587	100%	34 467

* Data from two centers were only available in an aggregated form which is unsuitable for breakdown in tabular format

eBOX. Collaborators.

Katrin Moritz¹, Dr. rer. biol. hum. Christopher Schulze¹, Dr. rer. biol. hum. Julia Zahn¹,

Dr. rer. biol. hum. Gabriele Ahne¹, PD Dr. med. Ines Marek¹, Eva Neumann², Dr. Melanie Schranz³, Prof. Dr. med. Freia de Bock³, Christine Gräf³, Sarah Leitzen⁴, Dr. rer. nat. Oliver Scholle⁵, Michelle Haaße⁶, Dr. rer. nat. Alexander Schnitzler⁶, Dr. rer. nat. Ghainsom Kom⁷, Nadine Steinkat⁷, Prof. Dr. rer. nat. Martin Schulz⁸, Dr. Gudrun Noleppa⁹, Dr. med. Katrin Bräutigam¹⁰,

Dr. med. Thomas Stammschulte¹⁰, Dr. med. Ursula Köberle¹⁰, Prof. Dr. rer. nat. Gerd Glaeske¹¹,

Dr. med. Hubert Radinger¹², PD Dr. med. Martina Pitzer¹³, Prof. Dr. med. Joachim Boos¹⁴,

Katrin Oelmann¹⁵, Dr. med. Kerstin Schindler¹⁶, Ilona Martens¹⁷, Birthe Herziger¹⁸,

Ruth Melinda Müller¹⁸, Dr. med. Malik Aydin¹⁹, Esma Durmaz¹⁹

 ¹ Department of Pediatric and Adolescent Medicine at the University Hospital of Erlangen

 ² Dr. Margarete Fischer-Bosch Institute of Clinical Pharmacology, Robert Bosch Society for Medical Research Stuttgart

 ³ Institute for Medical Biometry, Epidemiology and Information Technology, University Medicine at the Johannes Gutenberg University of Mainz

 ⁴ Federal Institute for Drugs and Medical Devices (BfArM), Research Department

 ⁵ Leibniz Institute for Prevention Research and Epidemiology – BIPS

 ⁶ University Hospital of Aachen, Department for Pediatric and Adolescent Medicine

 ⁷ TK Health Insurance Fund, Hamburg

 ⁸ Federal Union of German Associations of Pharmacists (ABDA)

 ⁹ German Society for Pediatric and Adolescent Medicine (DGKJ)

¹⁰ Drug Commission of the German Medical Association (AKdÄ)

¹¹ University of Bremen, Research Center on Inequality and Social Policy

¹² Joint Practice for Pediatric and Adolescent Medicine Bonn

¹³ Vitos Pediatric and Adolescent Clinic for Mental Health Eltville

¹⁴ University Hospital of Münster, Department for Pediatric and Adolescent Medicine

¹⁵ Chemnitz Hospital

¹⁶ Helios Hospital Krefeld, Center for Pediatric and Adolescent Medicine

¹⁷ University Department for Pediatric and Adolescent Medicine, Johannes Wesling Hospital Minden, Ruhr University of Bochum

¹⁸ University Medical Center of Rostock, Clinic for Children and Adolescents

¹⁹ Helios University Hospital of Wuppertal, Center for Pediatric and Adolescent Medicine