Association of the STOPP criteria
v2 and the Fall-Risk-Increasing
Drugs list with falls in older
hospitalized patients
Master thesis Medical Informatics
Project information
Author
Kimberly (Kimmy) Raven, BSc. Student number: 11025913
k.e.raven@amsterdamumc.nl
Mentor
Birgit Damoiseaux-Volman, MSc. PhD student SCOPE study
Department of Medical Informatics, Amsterdam University Medical Centers – location AMC
b.a.damoiseaux@amsterdamumc.nl
Tutor
Dr. Danielle Sent – Assistant professor
Department of Medical Informatics, Amsterdam University Medical Centers – location AMC
d.sent@amsterdamumc.nl
Location
Department of Medical Informatics, Amsterdam University Medical Centers – location AMC Meibergdreef 9,
1105 AZ Amsterdam
Period
3 Table of contents
Table of contents
English summary ... 4 Nederlandse samenvatting ... 5 1. Introduction ... 6 1.1 Objective ... 7 2. Methods ... 8 2.1. STOPPs ... 8 2.2. FRIDs ... 8 2.3. Selection of falls ... 9 2.4. Data retrieval ... 10 2.5. Data analysis ... 10 3. Results ... 12 3.1. STOPPs ... 12 3.2. FRIDs ... 12 3.3. Falls ... 13Univariate logistic regression ... 15
Propensity score matching... 16
4. Discussion ... 19
5. Acknowledgement ... 22
6. References ... 23
Appendix 1. Abbreviations ... 26
Appendix 2. Selection process of study population ... 27
Appendix 3. STOPPs ... 28
Appendix 4. Characteristics of the STOPPs ... 33
Appendix 5. Selected FRIDs ... 36
Appendix 6. Characteristics of FRIDs ... 37
Appendix 7. Search queries used in CTcue ... 38
Appendix 8. Regular expressions for problemlist search ... 40
Appendix 9: Characteristics of the study population ... 41
Appendix 10. Characteristics of the population with and without STOPPs ... 43
Appendix 11. Characteristics of the population with and without FRIDs ... 45
English summary
Objective
Falling amongst older persons is a growing problem, also during hospitalization. This is due to its high prevalence and related fatal and nonfatal injuries. The aim of this study is to identify the association between the STOPP criteria v2 and the FRIDs list on falling in older hospitalized patients using a large dataset of routinely collected, structured and unstructured EHR data of a cohort.
Subjects
Hospitalized patients of 70 years and older with a hospitalization duration of at least 24 hours, admitted between November 2015 up to November 2019.
Methods
A large dataset of hospitalizations of a university hospital was derived from the electronic health records. Identification of STOPP violations was performed by means of the Dutch STOPP-criteria v2. Identification of FRID administrations was performed by means of a European FRIDs list. Falls were identified by searching free-text nursing and physician notes and the list of running diagnoses and complications (known as the problemlist). Univariate logistic regression and doubly-robust propensity score matching was performed to analyze the risk that STOPP violations and FRID administrations pose on the occurrence of falls.
Results
Data included 16,823 hospital admissions for 11,354 patients. The median age of the population was 76, and 51.8% was male. In 56.7% of the hospitalizations one or more STOPP violations were
registered. FRID administrations were registered in 82.9% of the hospitalizations. One or more falls occurred in 257 (1.5%) of the hospitalizations. We found that, among others, male gender, age, length of stay, fall history, and an elevated delirium risk score (DOSS >3) increase the risk of falling in older hospitalized patients. We also found a decrease in falls in case a single FRID was administered (OR: 0.0002 (0.0001 – 0.0003)) or a single STOPP was violated (OR: 0.0042 (0.0035 – 0.0050)). However, an increase in fall risk was found when 8 or more different FRIDs were administered (OR: 4.0606 (3.9245 – 4.2047) for each additional FRID administration) or 5 or more individual STOPPs were violated (OR: 4.4823 (4.3395 – 4.6325) for each additional STOPP violated).
Conclusion
Overall, a prevalence of 82.9% for FRID administrations and a prevalence of 56.7% for STOPP violations was found. Furthermore, we found both a FRID administration and a STOPP violation in 54.9% of the hospitalizations. Both FRID administrations and STOPP violations showed a reduced risk of falls. However, the number of FRIDs administered and STOPPs violated does seem to increase the risk of falling in older hospitalized patients. These results would suggest that not the FRID or STOPPs itself are harmful, but the total amount are. Further research, with a larger number of fallers is needed to pose more trustworthy results. Furthermore, future research should also look at individual STOPP criteria or FRIDs medications to determine the association for these individual medications of STOPP criteria.
Keywords
5 Nederlandse samenvatting
Nederlandse samenvatting
Doelstelling
Vallen onder ouderen is een groeiend probleem vanwege hoge prevalentie en gerelateerde fatale en niet-fatale verwondingen. Het doel van dit onderzoek is het identificeren van de correlatie tussen de schendingen van de STOPP criteria v2 en de toedieningen van medicatie van de FRIDs lijst op vallen bij oudere, in het ziekenhuis opgenomen, patiënten. Hierbij wordt gebruik wordt gemaakt van een grote dataset van gestructureerde en ongestructureerde data uit het elektronisch patiëntendossier.
Populatie
In het ziekenhuis opgenomen ouderen van 70 jaar en ouder, met een opnameduur van ten minste 24 uur, opgenomen tussen November 2015 en November 2019.
Methode
Voor dit onderzoek is een grote dataset van ziekenhuisopnamen uit het elektronisch patiënten dossier gebruikt. Schendingen van STOPP criteria zijn geïdentificeerd door middel van de
Nederlandse STOPP criteria v2. De identificatie van toedieningen van FRID medicatie is gedaan door middel van een Europese FRIDs lijst. Vallen zijn geïdentificeerd door middel van het doorzoeken van vrij-tekst data (verpleegkundige notities en notities van de arts) en de lijst met lopende diagnoses en complicaties (probleemlijst). Univariate logistische regressie en “doubly robust” propensity score matching is uitgevoerd om het risico van schendingen van STOPPS en toedieningen van FRIDs op het plaatsvinden van vallen bij ouderen in kaart te brengen.
Resultaten
De data bevatte 16.823 ziekenhuisopnamen van 11.354 patiënten. De mediaan van de leeftijd was 76 jaar en 51,8% van de populatie was man. Tijdens 56,7% van de ziekenhuisopnamen werd een of meerdere STOPPs geschonden. Een of meer FRIDs werden toegediend in 82,9% van de opnames. In 257 (1,5%) van de opnames hebben een of meerdere vallen plaats gevonden. We vonden dat onder andere het mannelijk geslacht, leeftijd, opnameduur, een geschiedenis van vallen, en een verhoogde delier screening score (DOSS score (>3)), de kans op een val verhogen. Daarnaast vonden we een negatieve associatie tussen zowel het toedienen van een FRID en vallen (OR: 0.0002 (0.0001 – 0.0003)) als het schenden van een STOPP criteria en vallen (OR: 0.0042 (0.0035 – 0.0050)). Echter bleek een verhoogd risico voor vallen wel te bestaan bij de schending van 5 of meer verschillende STOPP criteria (OR: 4.4823 (4.3395 – 4.6325)), en de toediening van 8 of meer verschillende FRIDs (OR: 4.0606 (3.9245 – 4.2047)).
Conclusie
In onze studie vonden we een prevalentie van 82.9% voor FRID toedieningen en een prevalentie van 53.7% voor STOPP schendingen. In 54.9% van de opnames werden zowel een STOPP criteria
geschonden als een FRID medicatie toegediend. Zowel het schenden van een STOPP als het toedienen van een FRID bleek een verlagend effect te hebben met het voorkomen van vallen bij ouderen. Echter lijk wel het aantal toegediende verschillende FRIDs en verschillende geschonden STOPP criteria van belang te zijn. Verder onderzoek, met een grotere populatie gevallen patiënten is nodig om de resultaten betrouwbaarder te maken. Verder zou vervolgonderzoek zich moeten richten op het effect van schendingen van individuele STOPP criteria en toedieningen van verschillende medicamenten uit de FRIDs lijst.
Trefwoorden
1. Introduction
Falling amongst older persons is a growing problem, due to its high prevalence and related fatal and nonfatal injuries [1]. According to the World Health Organization (WHO), in the United States, 20-30% of older persons who fell suffer from serious injuries [2]. Most recent numbers, from 2014, show that in the United States approximately 29% of the persons aged 65 years and older has fallen at least once [1]. Furthermore, approximately 6% of the total population of older persons were treated in emergency departments for injuries caused by a fall [1, 3, 4]. For almost 29% of these persons this resulted in hospitalization, with average costs of $30,550 per hospitalization [5]. In The Netherlands, falling amongst older persons is a problem as well. In 2018, 108,000 elderly of 65 years and older (approx. 3.4% of the older persons), visited the emergency department due to a fall incident [6-8]. Falls can occur in private situations, but also during hospitalization. According to a study by Lakhan et al., approximately 6.4% of patients aged 70 and older, suffer from one or more falls during hospital stay [9]. Of these falls 30-40% result in injury, and approximately 4-6% has serious harm as a result. Of all inpatient falls that result in moderate to severe injury, 60% occurs in persons aged 70 years and older [10].
Risk factors for falls are, for example, gender, use of certain medication types, history of falls, balance disorder or cognitive decline [11-13]. To reduce the risk of falling during hospitalization, guidelines have been developed to identify and eliminate these risk factors, and to prevent falls, the fall risk for patients is estimated [14, 15]. The area in which probably the greatest risk reduction on a patient’s fall risk can be accomplished by the nurse or physician, is in the awareness of medication prescription [16]. For the identification of risk drugs, lists identifying potentially inappropriate prescribing (PIP) exist [17]. PIPs can be divided into potentially inappropriate medications (PIMs), containing for example drug-drug interactions, and potential prescribing omissions (PPOs), containing advice on, for example, prescribing a stomach protector when prescribing potentially harmful medication such as NSAIDs.
An example of such a list that identifies PIPs, is the Fall-Risk-Increasing Drugs (FRIDs) list. The FRIDs list is a list of medications that may increase the risk of falling in older people and advices to stop administration of these medication in patients with a high risk of falling. This list only focuses on PIMs. According to a study by Van der Velde et al., there is a significant difference between use of FRIDs and no use of FRIDs on the occurrence of falls (89% vs 76%) [18].
However, also more general lists identifying PIPs exist. These lists are not focused on a specific type of medication. Examples of these more general PIP lists are the Beers criteria of the American Geriatrics Society (AGS), which focuses on PIMs that should be avoided by older adults, and the STOPP/START criteria, containing both PIMs and PPOs, from Ireland [19, 20]. It is stated that the STOPP/START criteria fit best in the Dutch healthcare system when compared to the Beers criteria [21].
As mentioned before, another example of a list identifying PIPs, are the STOPP/START criteria. In these criteria, the STOPP criteria consist of rules concerning PIMs and the START criteria focus on PPOs [20]. The STOPP/START screening tool is designed to enable the prescribing physician to appraise an older patient’s prescription drugs, but in the context of the current diagnosis of the patient [22]. The STOPP criteria have a designated section for violations that can cause falls (section K). Experts on geriatric pharmacotherapy and/or falls stated that other sections also contain drugs that can be related to falls [23].
Both the STOPP criteria and the FRIDs list show the role of medication on falling. However, currently no study has compared the association of the violation of STOPP criteria and the administration of medications of the FRIDs list on the occurrence of falls. Studies after the association of either the STOPP criteria or FRIDs on falls exist, but these are often conducted in a different setting, for example in outpatient clinics [24, 25].
7 1. Introduction
1.1 Objective
This study focusses on identifying the possible role of the risk drugs as mentioned in the STOPP criteria and FRIDs in older patients on falling during a hospitalization with a duration longer than 24 hours. Risk drugs will be identified by means of the STOPP criteria v2 and the FRIDs list. The objective of this study is to identify the association between the STOPP criteria v2 and the FRIDs list on the occurrence of falls in older hospitalized patients using a large dataset of routinely collected, structured and unstructured EHR data of a cohort.
This thesis is organized as follows: The different methodologies used in this study and further explanation on them can be found in section 2. Section 3 shows the results of the study, which are discussed in section 4. In section 4 also the conclusion of this study can be found.
2. Methods
The study used routinely collected data from the hospital electronic health Record (EHR) to answer the defined retrospective research question. Hospital admissions with an admission date from November 1st, 2015 to November 1st, 2019 were selected. This resulted in data of more than 24,330 unique patients, with a total of 26,040 unique hospitalizations. Data include gender, age at
admission, medication prescriptions, dosage, prescription start and end dates, diagnoses, lab results, fall risk scores (Johns Hopkins Fall Risk Assessment Tool (JHFRAT) [26]) and delirium risk scores (Delirium Observation Screening Scale (DOSS) [27]), medication administrations, Anatomical
Therapeutic Chemical Classification (ATC) codes and diagnostic test results. All patients were aged 70 years or older at time of admission to the hospital with a hospital stay longer than 24 hours, to exclude day admissions. Only entries from clinical departments were included; admissions on short stay and the emergency department were not included. Furthermore, we only included records with no missing values in medication administrations. We excluded medication administrations with “planned” as administration status. This status indicates that the medication has not been
administered, while we only wanted to include administered medications. The selection process of the study population can be found in Appendix 2.
2.1. STOPPs
In this study, we used the Dutch version of the STOPP v2 criteria. An example of a criterion that can be found in the STOPP criteria is the combination of a beta blocker in combination with verapamil or diltiazem (STOPP B3). For the coding of the STOPP criteria the conversion of the criteria as proposed by Huibers et al. was used [28]. We altered some STOPP criteria by combining the Dutch and English version, of excluded criteria when no data was available. The different STOPP criteria, together with the alterations and possible reasons for exclusion can be found in Appendix 3. Even though the STOPP criteria have a designated section for falls (section K), we assessed the influence of violations of all STOPP criteria in this study, which resulted in the inclusion of 68 different STOPP criteria. In case a criterion focused on the interaction of two different medications, it was deemed a violation when the second drug was administered within 24 hours of the administration of the first drug. An example, using the previously mentioned STOPP B3, would be the administration of verapamil within 24 hours after the administration of a beta blocker. The selection of a timeframe of 24 hours was based upon the assumption that it is expected that within these medications mostly a short-term effect can be found. Per hospitalization, we registered a violation of each individual STOPP criterion only once.
2.2. FRIDs
Currently no European FRIDs list is available [29]. For this study, we used the FRIDs list from Seppala et al. [30]. This list contains for example diuretics and benzodiazepines. All of the medication (sub)classes found in the data were classified and coded using ATC codes by LR and checked by KE. For finding the correct ATC codes per medication class/type, we used the WHO ATC index and the CAREFREE consortium’s data harmonization guide [31, 32]. The final FRIDs list with the medication classes and ATC codes used in this study can be found in Appendix 5, and contains 21 different medications and medication classes. We registered the administration per unique FRID medication only once per hospitalization. Administrations were not registered per ATC class, for example C03 (diuretics), but per unique medication, for example C03AA05 (polythiazide).
9 2. Methods
2.3. Selection of falls
Electronic health records (EHRs) contain defined datasets of structured data, such as diagnoses, medications and demographics [33]. Furthermore, unstructured data, such as clinical notes, are available [34]. A study in primary care by Baus et al., found that more than 30% of all occurred falls were only registered in free-text, and emphasizes the usefulness of free-text search [35]. We assume similar results in secondary care. We therefore aimed to search both in structured as well as in unstructured data from the EHR.
For the free-text analysis CTcue version 2.0.10 was used [36]. CTcue searches in EHR data, both structured and unstructured, for matches on an entered query. Patients that possibly match the criteria, for example a certain age and a diagnosis, are returned. The user can view and check the returned patients, and select the patients that match the search criteria. We created two different search queries to identify falls. These can be found in Appendix 7. The free-text notes we included in our search, were nursing and physician notes taken at, during, and after admission.
Out of the patients that matched the query according to CTcue, falls were manually identified by KR and DS. A third reviewer, BD, checked patients for whom it was uncertain whether a fall occurred during hospitalization. The identification of falls was performed according the WHO definition for falls: “A fall is an event which results in a person coming to rest inadvertently on the ground or floor or other lower level” [37]. Patients without a fall were not further indexed for this part of the analysis. For all patients with a match, the date and time of the recording of a fall were manually extracted from the hit that was provided by CTCue. In case an exact time of fall was reported, this was selected to use, combined with the date on which the fall was reported. In case no exact time was reported in the free-text, we used the timestamp of the report or note as provided by CTcue. This date and time provide the moment the note was created. This data was connected to the original dataset. Please note that a larger set of fall-data was identified, and for this data the fall dates were transcribed. Unfortunately, this was not available by the end of underlying research and not used for the analysis.
We also searched the list with running diagnoses and complications (the ‘problemlist’) for falls that occurred as a complication during hospitalization. This was performed by means of two regular expressions. The first expression selected all entries that contained “fall” or “tripped”. The second expression excluded all irrelevant terms, for example “fall risk” and “tendency to fall". We found that falls recorded in the problemlist were coded as “Complication XXX fall” with the specific department on the XXX, for example INT for internal medicine. Collapse was left out of this search, since the occurrence of a fall due to a collapse could not be determined due to the lack of context provided. A collapse does not directly indicate a fall. The regular expressions we used can be found in Appendix 8. All entries in the problemlist get a default timestamp of 00:00:00, which made the moment of fall less accurate compared to the falls identified with CTcue. Due to this inaccuracy, we preferred the use of a fall recoded in CTcue over the use of a fall recorded in the problemlist, in case we identified a fall in both searches during the same hospitalization.
Out of all falls, we selected only one fall per hospitalization. Furthermore, not all falls identified in the free-text data could be linked to a hospitalization in the dataset.
2.4. Data retrieval
For each hospitalization, it was identified whether a FRID medication was administered, a STOPP criterion was violated, and a fall had occurred. In case a fall occurred during hospitalization, a FRID administration or STOPP violation were only recorded as a FRID or a STOPP in case the
administration occurred before the fall. For the characteristics of the different groups, gender is presented per hospitalization, since this is also how the logistic regression is performed. This means that one patient can have multiple hospitalizations, and that the gender of a patient may be counted several times.
We used the results of the Johns Hopkins Fall Risk Assessment Tool (JHFRAT) in several ways. We identified the number of patients that had one or more fall risk screenings during hospitalization, and the hospitalizations with one or more fall risk assessments with a medium (6-13) or high (> 13) fall risk score. Furthermore, we divided the JHFRAT into the different subcategories of age, fall history, toilet demand (elimination, bowel and urine), medications, PCE (patient care equipment), mobility, and cognition. We excluded age and medication scores from further analysis, since these were already present in our data and would be redundant. For each hospitalization it was registered whether the patient had a score higher than 0 for the remaining subcategories. This resulted in the recording of fall history (yes/no), risk toilet demand (yes/no), risk PCE (yes/no), mobility impairment (yes/no) and cognitive impairment (yes/no).
To gain insight in the influence of different diseases on falling, not only the number of diagnoses per hospitalization was computed. We also searched for diagnoses, either ICD-9 or ICD-10, and divided them in different classes, which can be found in Appendix 12. The ICD codes used were derived from the NICTIZ terminology [38].
2.5. Data analysis
For the different characteristics and the comparison between the groups, student’s t-test, Mann-Whitney U test or chi-square test were used where necessary. We performed univariate logistic regression to analyze the independent (crude) association between different the characteristics and the occurrence of falls. Medium and high fall risk were excluded from further analysis due to the overlap of the categories medication and age, that are included in the calculation of the fall risk score using the JHRFAT. Including the results of the JHFRAT would skew our outcomes, since we already included age and medications as separate categories.
To identify the association between STOPP violations and falls and FRID administrations and falls, propensity score matching was used. We preferred propensity score matching, in favor of
multivariate logistic regression. We chose this to ensure comparable groups in the analysis and more trustworthy and accurate outcomes due to the comparableness of both groups after matching since the matching process simulates a randomized controlled trial (RCT). Variables used for the
calculation of the propensity score were significant in the univariate logistic regression for falls. The propensity score was used to reduce the influence of selection and treatment allocation bias, and calculated by means of a multivariate logistic regression. For the calculation of the propensity score, we only included variables with a significant outcome in the univariate logistic regression. Greedy matching, or 1-nearest-neighbor (1:1), was used for matching. Furthermore, we used the R Matching package (Multivariate and Propensity Score Matching Software with Automated Balance
11 2. Methods Matching was performed with replacement and a caliper of 0.2 standard deviations. This caliper indicates how large the difference between two subjects may be, to be still matched to each other. We chose to perform matching with replacement, since the population receiving a FRID is much larger than the population which did not receive a FRID. This is similar for the population with STOPP violations when compared to the group without STOPP violations. Replacement was needed to match all hospitalizations with a FRID administration or STOPP violation to a hospitalization without a FRID administration or STOPP violation.
After matching, the standardized mean difference (SMD) of all variables in the data was checked to see if the matching between the population with and without STOPP violation or FRID administration was performed properly. A variable with an SMD > 0.2 was deemed a bad match. We also checked the balance by printing the propensity scores in a back to back histogram. In case the plot looked symmetrical or mirrored, the matching process was performed correctly. The output of the matching was used in another logistic model on falls to ensure doubly robust outcomes. In case we found any variables with an SMD > 0.2 after matching, these variables were also entered in the doubly robust logistic regression to adjust for this discrepancy in matching.
A p-value < 0.05 was considered significant. For this study, R version 3.6.1. “Action of the Toes” and CTcue version 2.0.10 have been used. The following R packages were used: readr, dplyr, stringr, plyr, gdata, Rcpp, rlang, DBI, odbc, tidyr, lubridate, ggplot2, lme4, TableOne, Matching.
3. Results
The selection of hospitalizations as mentioned earlier resulted in a total of 11,354 patients with a total of 16,823 hospitalizations. 52.4% of the population was male and the median (IQR) age of the population was 76 (72-81). Almost 88% of the subjects has had one or more fall risk assessments, and almost 45% had a medium fall risk (fall risk score 6 - 13). Furthermore, more than 68% of the population has one or more diagnosed cardiovascular diseases. In almost 83% of the hospitalizations one or more FRIDs was administered, and in almost 57% one or more STOPP criteria was violated. A violation of a STOPP and an administration of a FRID occurred in almost 55% of the hospitalizations. Further characteristics of the population can be found in Appendix 9.
3.1. STOPPs
Out of the total of 16,823 hospitalizations, in 9,545 hospitalizations one or more STOPP criteria were violated. In the hospitalizations with a STOPP violation, 57.8% had a medium fall risk, compared to 26.8% in the group with no STOPPS. Furthermore, in 56.8% of the hospitalizations with a STOPP, patients had mobility impairment issues, compared to 27.2% in the group without a STOPP violation. Another significant characteristic is the administration of a FRID during a hospitalization with a STOPP violation. This is almost 96.8%, compared to 64.6% in the hospitalizations without any STOPPs. Further characteristics of the population with and without STOPP violations can be found in Appendix 10.
The STOPP criterion with the highest prevalence in the population was the administration of benzodiazepines (criterion: STOPP K1), with a prevalence of 22.1%. This STOPP was violated 27,108 times in a total of 3,712 hospitalizations. The criterion with the second highest prevalence
concerned the administration of loop diuretic as first-line treatment for hypertension (criterion: STOPP B6), with a prevalence of 14.6%, and a total of 19,749 violations during 2,450 hospitalizations. The section of criteria with the highest prevalence is section B, criteria concerning the cardiovascular system, with a total prevalence of 39.33%. The prevalence per STOPP criterion can be found in Appendix 4.
3.2. FRIDs
In 13,944 of the 16,823 hospitalizations, one or more FRIDs were administered. When compared to the administration without any FRID administration, the median age is the same, namely 76 years. However, the median length of stay in the group with FRID administrations is more than two times longer than the median length of stay of the group without FRID administrations (4.99 vs. 1.98 days). Another large difference between the FRID and no FRID group can be found in the number of unique medications received (19 vs. 7). Furthermore, 47.5% of the admissions with FRIDs has a medium fall risk (fall risk score 6-13), compared to 29.4% in the group without FRID administrations. Further characteristics of the population with and without FRID administrations can be found in Appendix 11.
The most administered FRIDs were from the medication class opioids (ATC code N02A). Medications from this category were administered in 12,986 hospitalizations, which is 77.2% of the total number of hospitalizations. The second most administered FRIDs were diuretics (ATC code C03), with administrations in 10,149 hospitalizations. This is 60.3% of the total number of hospitalizations. Further information of the number of FRID administrations per FRID category as stated in Appendix 5, can be found in Appendix 6.
13 3. Results
3.3. Falls
Of the 487 patients with possible falls that were identified using CTcue, 223 patients were deemed a match after manual validation. A total of 261 falls were identified in these patients. However, we selected only one fall per hospitalization, and not all falls identified in the free-text could be linked to a hospitalization in the dataset. This means that the final number of hospitalizations with falls we identified using CTcue was 224 in 218 individual patients. The problemlist search identified falls in 72 hospitalizations in as many patients, of which 39 showed overlap with the results of the free-text search. This can be found in Table 1. In total free-text data of 2,455 patients was manually validated. Due to problems with data retrieval, only a small subset of the fall-data was used for further
analysis, hence the smaller number of results.
Table 1. The number of falls found, divided in total falls, falls identified in problemlist, falls identified in free-text, and the number of falls identified in both problemlist and free-text.
Total number of
admissions with falls Admissions with falls in problemlist Admissions with falls in free-text Falls in both problemlist and free-text
257 72 (30.2 %) 224 (85.1 %) 39 (15.3 %)
When compared to non-fallers, the proportion males in the faller group is significantly higher (52.3% vs. 59.1%). Note that these results are per hospitalization, which means that several patients may be counted double. The median age of this population was 77 (IQR: 73-82), compared to 76 (IQR: 72-81) in the group without falls. The length of stay was almost four times longer in the fall group when compared to the group without falls (median (IQR): 16.18 (8.39-28.96)) vs. (4.07 (2.00-8.03)). Furthermore, fallers showed a significant higher prevalence in mobility impairment (87.5 % vs. 43.3%), cognitive impairment (59.9% vs. 11.1%), and an elevated delirium score (66.5% vs. 12.4%). We also found significant differences in medium fall risk (65.4% vs. 44.1%) and high fall risk (60.3% vs 13.3%). Further characteristics on the distribution of fallers and non-fallers can be found in Table 2.
Table 2. Characteristics of the population with and without falls during hospitalization. IQR: interquartile range. The p-value shows whether the difference between the two groups is significant. Significant p-values have been marked with a star (*).
No fall (n = 16,566) One or more falls (n = 257) P-value
Gender (male), n (%)
Male 8,661 (52.3%) 152 (59.1%) 0.034*
Female 7,905 (47.7%) 105 (40.8%)
Age (years), median (IQR) 76.00 (72.00 - 81.00) 77.00 (73.00 - 82.00) 0.012*
Deceased during hospitalization, n (%) 826 (5.0%) 23 (8.9%) 0.006*
Length of stay (days), median (IQR) 4.07 (2.00 - 8.03) 16.18 (8.39 - 28.96) <0.001*
Unique medications, median (IQR) 16.00 (10.00 - 24.00) 24.00 (16.00 - 35.00) <0.001*
Number of diagnoses, median (IQR) 5.00 (3.00 - 7.00) 9.00 (6.00 - 12.00) <0.001*
Elevated delirium score (DOSS score ( >
3)), N (%) 2,058 (12.4%) 171 (66.5%) <0.001*
Johns Hopkins Fall Risk Assessment Tool
Fall risk assessment, n (%) 14,474 (87.4%) 253 (98.4%) <0.001*
Medium fall risk (6-13), n (%) 7,301 (44.1%) 168 (65.4%) <0.001*
High fall risk ( > 13), n (%) 2,195 (13.3%) 155 (60.3%) <0.001*
Fall history, n (%) 3,520 (21.2%) 183 (71.2%) <0.001*
Mobility impairment , n (%) 7,180 (43.3%) 225 (87.5%) <0.001*
Cognitive impairment, n (%) 1,834 (11.1%) 154 (59.9%) <0.001*
Risk toilet demand, n (%) 2,726 (16.5%) 151 (58.8%) <0.001*
Risk PCE, n (%) 7,203 (43.5%) 194 (75.5%) <0.001*
PIMs
FRID administration, n (%) 13,706 (82.7%) 92 (35.8%) <0.001*
Number of FRIDs, median (IQR) 3.00 (2.00 - 5.00) 5.00 (3.00 - 8.00) <0.001*
Number of STOPPs, median (IQR) 2.00 (0.00, 3.00) 4.00 (2.00, 6.00) <0.001*
ICD category, n (%)
Certain infectious and parasitic
diseases (ICD cat 1) 1,430 (8.6%) 58 (22.6%) <0.001*
Neoplasms (ICD cat 2) 4,180 (25.2%) 62 (24.1%) 0.739
Diseases of the blood and blood-forming organs and certain disorders involving the immune mechanism (ICD cat 3)
1,636 (9.9%) 53 (20.6%) <0.001*
Endocrine, nutritional and metabolic diseases (ICD cat 4), N (%)
6,214 (37.5%) 124 (48.2%) 0.001*
Mental and behavioral disorders
(ICD cat 5) 1,599 (9.7%) 127 (49.4%) <0.001*
Diseases of the nervous system
(ICD cat 6) 2,337 (14.1%) 80 (31.1%) <0.001*
Diseases of the senses (ICD cat
7+8) 678 (4.1%) 19 (7.4%) 0.013*
Diseases of the circulatory system
(ICD cat 9) 11,368 (68.6%) 202 (78.6%) 0.001*
Diseases of the respiratory system
(ICD cat 10) 3,420 (20.6%) 86 (33.5%) <0.001*
Diseases of the digestive system
(ICD cat 11) 2,552 (15.4%) 50 (19.5%) 0.090
Diseases of the skin and
subcutaneous tissue (ICD cat 12) 604 (3.6%) 24 (9.3%) <0.001*
Diseases of the musculoskeletal system and connective tissue (ICD cat 13)
1,651 (10.0%) 40 (15.6%) 0.004*
Diseases of the genitourinary
system (ICD cat 14) 3,927 (23.7%) 109 (42.4%) <0.001*
Pregnancy, childbirth and the
puerperium (ICD cat 15) 0 (0.0%) 0 (0.0%) <0.001*
Certain conditions originating in
the perinatal period (ICD cat 16) 0 (0.0%) 0 (0.0%) <0.001*
Congenital malformations, deformations and chromosomal abnormalities (ICD cat 17)
105 (0.6%) 1 (0.4%) 0.924
Symptoms, signs and abnormal clinical and laboratory findings, not elsewhere classified (ICD cat 18)
3,294 (19.9%) 121 (47.1%) <0.001*
Injury, poisoning and certain other consequences of external causes (ICD cat 19)
2,565 (15.5%) 93 (36.2%) <0.001*
External causes of morbidity and
mortality (ICD cat 20) 3,249 (19.6%) 117 (45.5%) <0.001*
Factors influencing health status and contact with health services (ICD cat 21)
9,010 (54.4%) 166 (64.6%) 0.001*
Codes for special purposes (ICD cat
15 3. Results
Univariate logistic regression
Our analysis shows, that having the male gender increases the risk of falling with a factor 1.3. Death during hospitalization increases the risk of falling with a factor of almost 2, and for each day the hospitalization takes, the fall risk increases by a bit more than one time. Furthermore, a fall risk assessment increases the risk on falling by more than factor nine. Similar numbers can be found for some of the sub-categories of the fall risk assessment: fall history (factor 9.4), mobility impairment (factor 9.1) and cognitive impairment (factor 11.6). Furthermore, an elevated delirium screening score (DOSS score > 3) increases the risk of falling with factor 14. We found that a diagnosis
concerning mental and behavioral disorders (ICD cat 5) increases the risk of falling with nearly factor nine. Remarkable outcomes can be found in the STOPP violation and FRID administration. Our results show that receiving a FRID lowers the chance of falling by 90%, and for STOPP violations the risk on falling is reduced by more than 70%. Further significant results of the univariate logistic regression can be found in Table 3.
Table 3. The significant results of the univariate logistic regression on falls. OR: Odds ratio; CI: confidence interval. Star (*): indicates significant outcomes
Independent variables Crude OR (95% CI) p value
Male gender 1.321 (1.030 - 1.701) 0.029*
Age 1.032 (1.011 - 1.053) 0.002*
Deceased during hospitalization 1.873 (1.182 - 2.826) 0.005*
Length of stay (days) 1.049 (1.044 - 1.055) <0.001*
Number of medications 1.046 (1.037 - 1.055) <0.001*
Number of diagnoses 1.204 (1.177 - 1.231) <0.001*
Elevated delirium score (DOSS
score (>3)) 14.017 (10.809 - 18.310) <0.001*
Johns Hopkins Fall Risk Assessment Tool
Fall risk assessment 9.142 (3.889 - 29.635) <0.001*
Fall history 9.165 (7.011 - 12.104) <0.001*
Mobility impairment 9.192 (6.440 - 13.575) <0.001*
Cognitive impairment 12.010 (9.330 - 15.521) <0.001*
Risk toilet demand 7.232 (5.631 - 9.322) <0.001*
Risk PCE 4.003 (3.027 - 5.367) <0.001* PIMs FRID administration 0.116 (0.090 - 0.150) <0.001* Number of FRIDs 1.277 (1.228 - 1.327) <0.001* STOPP violation 0.314 (0.238 - 0.410) <0.001* Number of STOPPs 1.462 (1.398 - 1.528) <0.001* ICD categories
Certain infectious and parasitic
diseases (ICD cat 1) 3.085 (2.273 - 4.125) <0.001*
Diseases of the blood and blood-forming organs and certain disorders involving the immune mechanism (ICD cat 3)
2.371 (1.729 - 3.195) <0.001*
Endocrine, nutritional and
metabolic diseases (ICD cat 4) 1.553 (1.213 - 1.988) <0.001*
Mental and behavioral
disorders (ICD cat 5) 9.144 (7.119 - 11.743) <0.001*
Diseases of the nervous system
(ICD cat 6) 2.752 (2.096 - 3.582) <0.001*
Diseases of the senses (ICD cat
7+8) 1.871 (1.127 - 2.922) 0.010*
Diseases of the circulatory
Diseases of the respiratory
system (ICD cat 10) 1.933 (1.482 - 2.504) <0.001*
Diseases of the skin and subcutaneous tissue (ICD cat 12)
2.722 (1.731 - 4.088) <0.001*
Diseases of the musculoskeletal system and connective tissue (ICD cat 13)
1.665 (1.168 - 2.314) 0.003*
Diseases of the genitourinary
system (ICD cat 14) 2.370 (1.843 - 3.040) <0.001*
Symptoms, signs and abnormal clinical and laboratory findings, not elsewhere classified (ICD cat 18)
3.585 (2.795 - 4.592) <0.001*
Injury, poisoning and certain other consequences of external causes (ICD cat 19)
3.095 (2.384 - 3.996) <0.001*
External causes of morbidity
and mortality (ICD cat 20) 3.425 (2.669 - 4.389) <0.001*
Factors influencing health status and contact with health services (ICD cat 21)
1.530 (1.186 - 1.986) 0.001*
Propensity score matching
The covariates used for the calculation of the propensity score can be found in Table 3. No missing values were present in these variables. After matching on propensity score, none of the variables used in the calculation of the propensity score had an SMD > 0.2. Figure 1 shows the distribution of the propensity score matching between the population with and without STOPP violations. Figure 2 shows the distribution of the propensity score matching between the population with and without FRID violations. Since both figures are symmetrical, we can state that the matching process resulted in two comparable groups for both FRID and no FRID and STOPP and no STOPP. Matching was performed in a greedy manner; k nearest neighbor was user with n=1 and with replacement. For doubly robust outcomes another logistic regression was performed.
17 3. Results
Figure 1. Visualization of the matching of propensity scores between the STOPP and no STOPP group. Left: population without STOPP violation; right: population with STOPP violation.
Figure 2. Visualization of the matching of propensity scores between the FRID and no FRID group. Left: population without FRID administration; right: population with FRID administration.
To identify the role of the number of STOPPS violated of FRIDs administered, the “number of FRIDs” and “number of STOPPs” variables were changed to “additional FRID” and “additional STOPP”. In this case, FRID/STOPP indicated the administration of one FRID or the violation of one STOPP. Each additional administration of another FRID or violation of another STOPP per hospitalization is counted in this variable. With only one FRID administration, additional FRID is 0, with 2 FRID
administration, additional FRID is 1. An additional FRID or STOPP is not the second administration or violation of the same FRID or STOPP, but the administration or violation of another FRID or STOPP. For example, it is not twice the administration of analides, but the administration of another FRID medication or violation of another STOPP criterion, for example the administration of biperiden. The FRID and additional FRID variable (or STOPP and additional STOPP variable) were entered in the doubly robust logistic regression.
The violation of one STOPP showed an OR (95% CI) of 0.0042 (0.0035 – 0.0050) with an OR (95% CI) of 4.4823 (4.3395 – 4.6325) for each additional violated STOPP criterion other than the first violated STOPP. An OR (95% CI) of 0.0002 (0.0001 – 0.0003) was found for the administration of a FRID, with an OR (95% CI) of 4.0606 (3.9245 – 4.2047) for each additional administered FRID medication other than the already administered FRID. The results of the propensity score matching can be found in Table 4 for STOPP violations and Table 5 for FRID administrations.
Table 4. The results of the doubly robust multivariate regression for STOPP violations after propensity score matching. OR: Odds Ratio; CI: confidence interval. Additional STOPP indicates every additional, unique STOPP criterion violated, when the total of unique individual STOPP violations is higher than 1. Significant p values have been marked with a star (*).
Adjusted OR (95% CI) P value
One STOPP 0.0042 (0.0035 - 0.0050) < 0.001*
Additional STOPP 4.4823 (4.3395 - 4.6325) < 0.001*
Table 5. The results of the doubly robust multivariate regression for FRID administrations after propensity score matching. OR: odds ratio; CI: confidence interval. Additional FRID indicates every individual unique FRID administered, when the total of unique individual FRID administrations is higher than 1. Significant p values have been marked with a star (*).
Adjusted OR (95% CI) P value
One FRID 0.0002 (0.0001 - 0.0003) < 0.001*
19 4. Discussion
4. Discussion
This study, concerning the association of STOPP violations and FRID administrations on the
occurrence of falls in older patients, shows a significant association between the administration of a single FRID and the violation of a single STOPP. However our outcomes show a decrease in fall risk when a single FRID is administered of a single STOPP is violated during hospitalization. Univariate logistic regressions showed that male gender, age, deceased during hospitalization, length of stay, number of medications, number of diagnoses, fall risk assessment, fall history, mobility impairment, cognitive impairment, risk toilet demand, risk PCE, number of FRIDs, number of STOPPs, and an elevated delirium risk score (>3) show an increase in fall risk. Diseases from 15 different ICD diagnosis classes also seem to have an association with increased fall risk.
When compared to the structured data in the problemlist, more than 85% of the falls were identified with the free-text search. Our findings of the propensity score matching, show a decrease in fall risk in case of one STOPP violation and one FRID administration (respectively OR: 0.0042 (0.0035 – 0.0050) and OR: 0.0002 (0.0001 – 0.0003)). These findings do not support out hypothesis, which stated that STOPP violations and FRID administrations both would increase the risk of falling in elderly persons during hospitalization. However, these results are for administration of one FRID or violation of one STOPP only. In case a patient receives more than one unique FRID administered or one unique STOPP violated, the baseline risk (for only one administration) is increased by a factor 4.06 per additional FRID administration and a factor 4.48 per additional STOPP violation. For the FRIDs, this indicates that the administration of 8 or more FRIDs shows an increase of the fall risk in older hospitalized patients. For the STOPP criteria, the violation of 5 or more individual STOPPs show an increase in the fall risk.
Our findings concerning falling in older hospitalized patients, partially meet the findings of other studies. Compared to Baus et al., we identified a lot more falls in the free-text than in the structured data (85.1% vs. 30.2%), where Baus et al. identified 30% of the total number of falls in free-text. However, this could be explained by a difference in setting, since our study was performed in secondary care and the study of Baus et al. in primary care [35]. Furthermore, we found that in approximately 1.5% of the older hospitalized patients one or more falls occurred, which is much lower than the 6.4% found by Lakhan et al [9]. Our lower percentage of identified falls may be caused by the use of only a small part of the data for further analysis, due to data-processing and availability issues we encountered.
There seems to be a lack of consensus regarding the influence of both gender and age on fall risk. Our findings show significant increased risk of falling based on male gender (OR: 1.321 (1.030 - 1.701)). However, several studies, similar to ours, show no significant influence of gender or pose female gender as significant risk factor [39, 40]. For age, we also found a significant influence (OR: 1.032 (1.011 - 1.053)). This finding is supported by Mazur et al., but Najafpur et al. shows no
significant influence of age [39, 41]. A history of falls shows a significant increase in fall risk, which is supported by Mazur et al. and Mecconi et al. [41, 42]. The latter also states that cognitive
impairment increases, which also corresponds to our findings.
We found a decrease in risk between the administration of a single FRID and increased falling in older hospitalized patients, with an OR (95% CI) of 0.0002 (0.0001 – 0.0003). This is in contradiction with the findings of Van der Velde et al, which reported a difference of 89% vs 76% in use of FRIDs in fallers and non-fallers. This difference in findings could be caused in the significant difference in use of FRIDs of 35.8 vs 82.7% between fallers and non-fallers. This lower percentage of FRIDs use in fallers could be caused by our selection of FRIDs only administered before a fall to have influence on the occurred fall. Furthermore, this could be caused by the relatively low number of fallers when compared to the very large number of patients with one or more FRIDs.
Another study by Van der Velde et al. found a larger number in drugs, together with a larger number of FRIDs in the faller group when compared to the non-faller group [25]. The population with falls also had a higher fall incidence baseline. FRID withdrawal or dose reduction in the faller group showed a significant reduction in falls, which indicates that FRIDs do increase the risk of the occurrence of falls. However, Van der Velde et al. only used patients from the outpatient clinic and diagnostic daycenter, and falls were reported by the patients themselves. Furthermore, the
population was relatively small (n=139), and only patients with one or more falls in the previous year and who could walk at least 10 meters without walking aid, were included. This could possibly explain the difference in results. However, despite we found a decrease in fall risk when a single FRID is administered or a single STOPP is violated, this risk is increased by 4.0606 times for every additional, different, FRID administered, which means that the fall risk is increased by the
administration of at least 8 different FRIDs (OR: 4.0606 (3.9245 – 4.2047) for each additional FRID). Our study found a decreased risk on the occurrence of a fall, in case a single STOPP vas violated (OR: 0.0042 (0.0035 – 0.0050)). Lawson et al. found no significant difference in PIM administration between recurrent fallers and non-fallers [43]. However, this study used the Beers’ criteria, and had a relatively small sample size (n=99). A randomized clinical trial by Frankenthal et al. found a
significant decrease in falls after 12 month follow-up after medication alteration using the STOPP criteria [44]. However, this study took place in a chronic geriatric facility. Masumoto et al., which identified PIMs by means of the STOPP criteria v2 in an outpatient primary care clinic, found that in the group with STOPP violations and polypharmacy (≥ 5 different medications) showed a significant increase in fall risk [24]. However, this increase in fall risk was not identified in the group with PIMs but without polypharmacy. Despite our results do not show an increase of fall risk due to the violation of a STOPP, we found an increase for each additional STOPP, similar to the results of the FRIDs. The baseline risk is increased by 4.48 times (OR: 4.4823 (4.3395 – 4.6325)) per additional, different STOPP violated. This indicates that 5 or more violations of unique STOPP criteria during one hospitalization, show a significant increase of the fall risk of older hospitalized patients.
To our knowledge, this study is the first to compare the STOPP criteria and the FRIDs list in the occurrence of falls. We identified a very large number of falls in the free-text nursing and physician notes (85.1%), compared to only 30.2% of the falls that were recorded in the problemlist. This large difference in number of falls identified shows the importance of a free-text search. In comparison to other studies, we used a large population. Furthermore, this study is one of the few conducted in a hospital setting and focused on falls during hospitalization. Using propensity score matching, we simulated an RCT in our data, and minimized the differences between the fallers and non-fallers, despite the large difference in size of the groups. Due to the nature of the data, and the use of medication administration data, we presume that most medications have been administered, which gives an advantage over pharmacological prescriptions. Due to data retrieval of hospitalizations of all clinical departments, we managed to get a large, varied population.
However, as mentioned before, the number of falls identified is much lower than other studies state. This can be explained by the small number of faller that was used for analysis, and the limitations the use of CTcue entails. The use of CTcue was the only way to gain access to the free-text notes in the EHR. However, we identified a difference in results between the two searches we performed, which implies that not all fallen patients have been identified. Furthermore, the functionality of CTcue is focused on the identification of patients for RCTs, and manual extraction of fall dates and times made it prone to transcription errors. Besides, CTcue only searches for terms literally, so typos and sentences with other word orders than the ones provided in the query are not identified as a hit.
21 4. Discussion Even though more than 3,000 patient records were manually validated, only a small section of this data could be used for further analysis. The additional data contained 335 patients with 420 falls which took place during hospitalization. We think that this would mean, that the total number of falls in the study could be more than doubled. Furthermore, we think that the low number of fallers (when compared to non-fallers) could cause the unexpected outcomes. Due to looking at the STOPP criteria and FRID medications as a whole, no statements can be made on the effect of an individual STOPP criterion or individual FRID medications on the occurrence of falls in older hospitalized patients.
The use of matching with replacement was caused by a large number of hospitalizations with a FRID administration or STOPP violation when compared to the control group. This was the case for both STOPP violations and FRID administrations. Due to working with replacement, the bias was lowered, since the best fitting control can be chosen for the treatment [45]. Furthermore, this increases the quality of matching. However, this also means that the variance increases, and in our case, that a lot of control subjects were re-used many times.
For future research we would like to advise to look at the role of individual FRIDs and STOPPs on the occurrence of falls. It is possible that a specific STOPP criterion or section (e.g. section K) would indeed play a role in the occurrence of falls. Furthermore, we would search another way of
identifying falls in free-text. In this study, we did not take dosages into account. However, it might be possible that dosages of medications could influence the occurrence of a fall. An expert team should determine the “normal” dosages for each risk drug in this specific setting and population to make this usable. Furthermore, future research can look at the possible influence of a stay on an operating room on the occurrence of falls. Finally, we think that looking at changes in medication
administration (i.e. medication is stopped or lowered in dose) during hospitalization and the association with falls could be interesting.
In conclusion, our study shows that both the violation of one STOPP and administration of one FRID decreases the chance of falling in older hospitalized patients. However, the number of administered FRIDs and violated STOPPs show an independent association which increases the risk of falling. This could mean that not the administration of FRIDs or violations STOPPs itself have an important role in falling, but the severity/number of violations does. Since several studies show other results, future research should increase the number of falls and divide the analysis for each STOPP criterion of individual FRID to identify the possible risk caused by more specific medications.
5. Acknowledgement
When I started with the Bachelor Medical Informatics in 2015 at the University of Amsterdam, I was not sure whether it would be the right choice. Luckily, it was, and after the bachelor I also started the Master Medical Informatics. During both the Bachelor and Master, I aimed to get the best possible results. Many times I was insecure and wondering whether I was working hard enough and delivering high enough quality. This feeling did not get any less during my SRP. Perhaps it became even more present, especially with the problems in data availability and COVID-19-related issues. For some reason I always have the feeling I am working too slow or not making enough progress. Finishing this thesis makes me feel proud of what I have accomplished.
I would not have been able to get this result without the aid and support of several people. First of all, I would like to thank my parents, for the confidence, patience and emotional support they gave when I felt like giving up. They had many great ideas and always took the time to listen to my ideas and gave feedback on whether it was usable for my thesis. Furthermore, during these strange Covid-19 times, they always kept me in mind, and adjusted their working hours or other things, so my surroundings would be the best possible working environment. Second, I would like to thank Lotte Ruchtie, for all her hard work in a short time span. Without her input and coding, this thesis would not be the same. I would also like to thank Birgit Damoiseaux-Volman, for the great feedback
(especially on my writing) and the quick answers on any questions I had. She even encouraged me to present during a research meeting, even though I am not that confident. Also her expertise on falling and medication helped me to gain insight in the subject and what was or was not possible. Finally, I would like to thank Danielle Sent, since she would make time for me any time I needed her for questions. Without her expertise on data science and data analysis the finishing of my thesis would most likely have taken longer. Furthermore, she helped me to make this thesis more understandable by her great feedback.
23 6. References
6. References
1. Bergen G, Stevens MR, Burns ER. Falls and Fall Injuries Among Adults Aged >/=65 Years - United States, 2014. MMWR Morb Mortal Wkly Rep. 2016;65(37):993-998.
2. World Health Organization. Fact sheets falls. [Internet]. 2018. Available from:
https://www.who.int/news-room/fact-sheets/detail/falls#:~:text=Key%20facts,greatest%20number%20of%20fatal%20falls. [Accessed June 15, 2020].
3. Colby S, Ortman J. Projections of the Size an Composition of the U.S. Population: 2014 to 2060. U.S. Department of Commerce, U.S. Census Bureau; 2015.
4. Prevention CfDCa. Injury Prevention and Control - Data and Statistics (WISQARS) [Internet]. 2020. Available from: https://www.cdc.gov/injury/wisqars/ [Accessed May 25 2020]. 5. Burns ER, Stevens JA, Lee R. The direct costs of fatal and non-fatal falls among older adults -
United States. J Safety Res. 2016;58:99-103.
6. Van Der Does H, Baan A, Panneman M. Privé- valongevallen bij ouderen. Cijfers valongevallen in de privésfeer 2018. Amsterdam: VeiligheidNL; 2018.
7. Eurostat. Population on 1st January by age, sex and type of projection. [Internet]. 2019 [updated January 27 2020]. Available from:
https://appsso.eurostat.ec.europa.eu/nui/show.do?dataset=proj_18np&lang=en [Accessed January 27 2020].
8. Volksgezondheidenzorg.info. Vergrijzing - Totaal aantal ouderen 2018. [Internet]. 2020. Available from: https://www.volksgezondheidenzorg.info/onderwerp/bevolking/cijfers-context/vergrijzing#node-totaal-aantal-ouderen [Accessed January 27 2020].
9. Lakhan P, Jones M, Wilson A, Courtney M, Hirdes J, Gray LC. A Prospective Cohort Study of Geriatric Syndromes Among Older Medical Patients Admitted to Acute Care Hospitals. J Am Geriatr Soc. 2011;59(11):2001-2008.
10. Hitcho EB, Krauss MJ, Birge S, Claiborne Dunagan W, Fischer I, Johnson S, et al. Characteristics and circumstances of falls in a hospital setting: a prospective analysis. J Gen Intern Med. 2004;19(7):732-739.
11. Ambrose AF, Paul G, Hausdorff JM. Risk factors for falls among older adults: a review of the literature. Maturitas. 2013;75(1):51-61.
12. Munch T, Harrison SL, Barrett-Connor E, Lane NE, Nevitt MC, Schousboe JT, et al. Pain and falls and fractures in community-dwelling older men. Age Ageing. 2015;44(6):973-979.
13. Ham AC, Swart KM, Enneman AW, van Dijk SC, Oliai Araghi S, van Wijngaarden JP, et al. Medication-related fall incidents in an older, ambulant population: the B-PROOF study. Drugs Aging. 2014;31(12):917-927.
14. Ferderatie Medisch Specialisten. Schatting valrisico ouderen in het ziekenhuis. [Internet]. 2017. Available from:
https://richtlijnendatabase.nl/richtlijn/preventie_van_valincidenten_bij_ouderen/schatting_v alrisico_ouderen_in_het_ziekenhuis.html#tab-content-starting-question [Accessed February 18 2020].
15. National Institute for Health and Care Excellence. Older patients at high risk of hospital falls. [Internet]. 2013. Available from: https://www.nice.org.uk/news/article/older-patients-at-high-risk-of-hospital-falls [Accessed February 18 2020].
16. Ölveczky D. How do I kep my elderly patients from falling? [Internet]. 2009. Available from:
https://www.the-hospitalist.org/hospitalist/article/123872/how-do-i-keep-my-elderly-patients-falling [Accessed February 18 2020].
17. O'Connor MN, Gallagher P, O'Mahony D. Inappropriate prescribing: criteria, detection and prevention. Drugs Aging. 2012;29(6):437-452.
18. van der Velde N, van den Meiracker AH, Pols HAP, Stricker BHC, van der Cammen TJM. Withdrawal of fall-risk-increasing drugs in older persons: Effect on tilt-table test outcomes. J Am Geriatr Soc. 2007;55(5):734-739.
19. By the American Geriatrics Society Beers Criteria Update Expert Panel. American Geriatrics Society 2019 Updated AGS Beers Criteria(R) for Potentially Inappropriate Medication Use in Older Adults. J Am Geriatr Soc. 2019;67(4):674-694.
20. O'Mahony D, O'Sullivan D, Byrne S, O'Connor MN, Ryan C, Gallagher P. STOPP/START criteria for potentially inappropriate prescribing in older people: version 2. Age and Ageing.
2015;44(2):213-218.
21. Vermeulen Windsant-van der Tweel A, Verduijn M, Derijks H, Van Marum R. Detectie van ondergeschikt medicatiegebruik bij ouderen - Worden de STOPP- en START- criteria de nieuwe stantaard. Ned Tijdschr Geneedkd. 2012;156(A5076):1-8.
22. Gallagher P, Ryan C, Byrne S, Kennedy J, O'Mahony D. STOPP (Screening Tool of Older Person's Prescriptions) and START (Screening Tool to Alert Doctors to Right Treatment). Consensus validation. International Journal of Clinical Pharmacology and Therapeutics. 2008;46(2):72-83. 23. Damoiseaux BA, Medlock S, Romijn JA, Karapinar F, Velde vdN, Henstra M, et al. Development of a clinical decision support system for medication review in older hospitalized patients with high risk of falls or delirium. 2020:(Unpublished paper).
24. Masumoto S, Sato M, Maeno T, Ichinohe Y, Maeno T. Potentially inappropriate medications with polypharmacy increase the risk of falls in older Japanese patients: 1-year prospective cohort study. Geriatr Gerontol Int. 2018;18(7):1064-1070.
25. van der Velde N, Stricker BH, Pols HA, van der Cammen TJ. Risk of falls after withdrawal of fall-risk-increasing drugs: a prospective cohort study. Br J Clin Pharmacol. 2007;63(2):232-237. 26. Johns Hopkins Medicine. Fall Risk Assessment - The Johns Hopkins Fall Risk Assessment Tool
(JHFRAT) [Internet]. 2007. Available from:
https://www.hopkinsmedicine.org/institute_nursing/models_tools/fall_risk.html [Accessed June 19 2020].
27. Nederlandse Vereniging van Ziekenhuizen (NVZ), Nederlandse Federatie van UMC's (NFU). Veiligheids agenda VSM zorg - DOSS Delirium Observatieschaal [Internet]. 2020. Available from: https://www.vmszorg.nl/praktijkvoorbeelden-en-tools/doss-delirium-observatieschaal/
[Accessed June 19 2020].
28. Huibers CJA, Sallevelt BTGM, de Groot DA, Boer MJ, van Campen JPCM, Davids CJ, et al. Conversion of STOPP/START version 2 into coded algorithms for software implementation: A multidisciplinary consensus procedure. International Journal of Medical Informatics.
2019;125:110-117.
29. Seppala LJ, van der Velde N, Masud T, Blain H, Petrovic M, van der Cammen TJ, et al. EuGMS Task and Finish group on Fall-Risk-Increasing Drugs (FRIDs): Position on Knowledge
Dissemination, Management, and Future Research. Eur Geriatr Med. 2019;10(2):275-283. 30. Seppala LJ, Petrovic M, Ryg J, Bahat G, Topinkova E, Szczerbinska K, et al. STOPPFall (Screening
Tool of Older Persons Prescriptions in older adults with high fall risk): a Delphi study by the EuGMS Task and Finish Group on Fall-Risk-Increasing Drugs.:(Submitted).
31. WHO Collaborating Centre for Drug Statistics Methodology. ATC/DDD Index 2020. [Internet]. 2020. Available from: https://www.whocc.no/atc_ddd_index/ [Accessed January 30 2020]. 32. CAREFREE consortium. Effective withdrawal of fall-risk-increasing drugs: a European approach
- Data harmonization guide. 2019.
33. Institute of Medicine (US) Committee on Data Standards for Patient Safety. Background. Key Capabilities of an Electronic Health Record System: Letter Report. Washington (DC)2003. 34. Ehrenstein V, Kharrazi H, Lehmann H, Taylor CO. Obtaining Data From Electronic Health
Records. In: Gliklich RE, Leavy MB, Dreyer NA, editors. Tools and Technologies for Registry Interoperability, Registries for Evaluating Patient Outcomes: A User's Guide, 3rd Edition, Addendum 2. AHRQ Methods for Effective Health Care. Rockville (MD)2019.
35. Baus A, Zullig K, Long D, Mullett C, Pollard C, Taylor H, et al. Developing Methods of Repurposing Electronic Health Record Data for Identification of Older Adults at Risk of Unintentional Falls. Perspect Health Inf Manag. 2016;13:1b.
25 6. References 36. CTCue. Find patients in an instant. [Internet]. 2020. Available from: https://ctcue.com/
[Accessed January 30 2020].
37. World Health Organization. Falls. [Internet]. 2020. Available from:
https://www.who.int/violence_injury_prevention/other_injury/falls/en/ [Accessed February 17 2020].
38. Rijksinstituut voor Volksgezondheid en Milieu, NICTIZ. ICD-10 NL 2014. [Internet]. 2014. Available from: https://terminologie.nictiz.nl/art-decor/claml?collection=icd10-nl-data
[Accessed June 12 2020].
39. Najafpour Z, Godarzi Z, Arab M, Yaseri M. Risk Factors for Falls in Hospital In-Patients: A Prospective Nested Case Control Study. Int J Health Policy Manag. 2019;8(5):300-306.
40. Falcao RMM, Costa K, Fernandes M, Pontes MLF, Vasconcelos JMB, Oliveira JDS. Risk of falls in hospitalized elderly people. Rev Gaucha Enferm. 2019;40(spe):e20180266.
41. Mazur K, Wilczynski K, Szewieczek J. Geriatric falls in the context of a hospital fall prevention program: delirium, low body mass index, and other risk factors. Clin Interv Aging.
2016;11:1253-1261.
42. Mecocci P, von Strauss E, Cherubini A, Ercolani S, Mariani E, Senin U, et al. Cognitive impairment is the major risk factor for development of geriatric syndromes during
hospitalization: results from the GIFA study. Dement Geriatr Cogn Disord. 2005;20(4):262-269. 43. Lawson K, Vinluan CM, Oganesyan A, Gonzalez EC, Loya A, Strate JJ. A retrospective analysis of
prescription medications as it correlates to falls for older adults. Pharm Pract-Spain. 2018;16(4).
44. Frankenthal D, Lerman Y, Kalendaryev E, Lerman Y. Intervention with the Screening Tool of Older Persons Potentially Inappropriate Prescriptions/Screening Tool to Alert Doctors to Right Treatment Criteria in Elderly Residents of a Chronic Geriatric Facility: A Randomized Clinical Trial. J Am Geriatr Soc. 2014;62(9):1658-1665.
45. Caliendo M, Kopeinig S. Some practical guidance for the implementation of propensity score matching. J Econ Surv. 2008;22(1):31-72.
Appendix 1. Abbreviations
ATC Anatomical Therapeutical Chemical Classification System
CI Confidence interval
DOSS Delirium Observation Screening Scale
EHR Electronic Health Record
FRID Fall-Risk-Increasing Drug
JHFRAT Johns Hopkins Fall Risk Assessment Tool
ICD International Classification of Diseases and Related Health Problems
OR Odds ratio
PIM Potentially Inappropriate Medication
PIP Potentially Inappropriate Prescription
PPO Potential Prescribing Omissions
RCT Randomized Controlled Trial
SMD Standardized Mean Difference
START Screening Tool to Alert to Right Treatment STOPP Screening Tool of Older Persons’ Prescriptions
27 Appendix 2. Selection process of study population
Appendix 2. Selection process of study population
Figure 3 shows a visualization of process of patient inclusion, together with the number of remaining
patients and hospitalizations and the number of excluded patients and hospitalizations.
Figure 3. Visualization of the selection process of the studied population. The number of unique admissions and unique patients are provided, along with the reason for exclusion.
Appendix 3. STOPPs
Table 6 shows which STOPPs exist, together with a short explanation of the criterion and possible alterations or exclusions of criteria from this study. A similar table, and further information on the selection of the STOPP criteria, can be found in Damoiseaux et al. [23].
Table 6. The existing STOPP criteria, together with the explanation of the criterion and possible alterations made or reason for exclusion..
STOPP
criterion Explanation of the criterion Notes
Section A: Indication of medication
A1 Any drug prescribed without an evidence-based clinical indication. Huibers et al. did not
provide a conversion for this STOPP section; this section was not coded or used in this study.
A2 Any drug prescribed beyond the recommended duration, where
treatment duration is well defined.
A3 Any duplicate drug class prescription e.g. two concurrent NSAIDs,
SSRIs, loop diuretics, ACE inhibitors, anticoagulants (optimization of monotherapy within a single drug class should be observed prior to considering a new agent).
Section B: Cardiovascular System
B1 Digoxin for heart failure with normal systolic ventricular function
(no clear evidence of benefit) No data for left ventricular ejection
fraction.
B2 Verapamil or diltiazem with NYHA Class III or IV heart failure (may
worsen heart failure).
B3 Beta-blocker in combination with verapamil or diltiazem (risk of
heart block). If one is administered within 24 hours after the
other.
B4 Beta blocker with bradycardia (< 50/min), type II heart block or
complete heart block (risk of complete heart block, asystole). No heartrate available in data, bradycardia left out.
B5 Amiodarone as first-line antiarrhythmic therapy in supraventricular
tachyarrhythmias (higher risk of side-effects than beta-blockers, digoxin, verapamil or diltiazem)
B6 Loop diuretic as first-line treatment for hypertension (safer, more
effective alternatives available). Dutch version does not contain “first-line”; no
heart failure coded as exclusion criterion.
B7 Loop diuretic for dependent ankle oedema without clinical,
biochemical evidence or radiological evidence of heart failure, liver failure, nephrotic syndrome or renal failure (leg elevation and /or compression hosiery usually more appropriate).
B8 Thiazide diuretic with current significant hypokalaemia (i.e. serum
K+ < 3.0 mmol/l), hyponatraemia (i.e. serum Na+ < 130 mmol/l) hypercalcaemia (i.e. corrected serum calcium > 2.65 mmol/l) or with a history of gout (hypokalaemia, hyponatraemia,
hypercalcaemia and gout can be precipitated by thiazide diuretic)
B10 Centrally-acting antihypertensives (e.g. methyldopa, clonidine,
moxonidine, rilmenidine, guanfacine), unless clear intolerance of, or lack of efficacy with, other classes of antihypertensives
(centrally-active antihypertensives are generally less well tolerated by older people than younger people)
B11 ACE inhibitors or Angiotensin Receptor Blockers in patients with
hyperkalaemia.
B12 Aldosterone antagonists (e.g. spironolactone, eplerenone) with
