Identifying and quantifying potentially problematic prescribing cascades in clinical practice: A mixed‐methods study

Abstract

Background

A prescribing cascade occurs when medication causes an adverse drug reaction (ADR) that leads to the prescription of additional medication. Prescribing cascades can cause excess medication burden, which is of particular concern in older adults. This study aims to identify and quantify potentially problematic prescribing cascades relevant for clinical practice.

Methods

A mixed‐methods study was conducted. First, prescribing cascades were identified through literature search. An expert panel (n = 16) of pharmacists and physicians assessed whether these prescribing cascades were potentially problematic. Next, a cohort study quantified potentially problematic prescribing cascades in adults using Dutch community pharmacy data for the period 2015–2020. Additionally, the influence of multiple medications potentially causing the same ADR was evaluated. Prescription sequence symmetry analysis was used to calculate adjusted sequence ratios (aSRs), adjusting for temporal prescribing trends. An aSR >1.0 indicates the occurrence of a prescribing cascade. In a subgroup analysis, aSRs were calculated for older adults.

Results

Seventy‐six prescribing cascades were identified in literature and three were provided by experts. Of these, 66 (83.5%) were considered potentially problematic. A significant positive aSR for the medication sequence was found for 41 (62.1%) of these prescribing cascades. The highest aSR was found for amiodarone potentially causing hypothyroidism treated with thyroid hormones (4.63 [95% confidence interval 4.40–4.85]), based on 565 incident users. The biggest population (n = 34,645) was found for angiotensin converting enzyme‐inhibitors potentially causing urinary tract infections treated with antibiotics. Regarding four potential ADRs, the aSRs were higher for people using multiple medications that cause the same ADR as compared to people using only one of those medications. Among older adults the aSRs remained significant for 37 prescribing cascades.

Conclusion

An overview was generated of potentially problematic prescribing cascades relevant for clinical practice. These results can support healthcare providers to intervene and reduce medication burden for older adults.

Key points

• From 79 distinct prescribing cascades identified from literature, a multidisciplinary expert panel assessed 66 prescribing cascades to be potentially problematic, defined as the benefits of the prescribing cascade potentially not outweighing the risks.

• For 41 of the 66 potentially problematic prescribing cascades, a statistically significant positive association for the medication sequence was found among adults, supporting their occurrence in primary care. Of these 41 prescribing cascades, 37 remained significant among older adults.

• The use of multiple medications that may cause the same adverse drug reaction can increase the likelihood of a prescribing cascade.

Why does this paper matter?

This article presents an overview of potentially problematic prescribing cascades that are likely to occur in clinical practice. This overview can assist healthcare providers in identifying and managing prescribing cascades. This is relevant to decrease medication burden in older adults. Furthermore, it illustrates that extra attention is needed when people use multiple medications that can cause the same adverse drug reaction (ADR), since this may increase the likelihood of problematic prescribing cascades.

INTRODUCTION

A prescribing cascade occurs when medication causes an adverse drug reaction (ADR) that leads to the prescription of additional medication. ¹ , ² This can be intentional to manage the ADR but this is not the case when the ADR is misinterpreted as a new condition. Although some intentional prescribing cascades can be appropriate or even recommended, many prescribing cascades are considered problematic. Prescribing cascades are considered problematic when the benefits of prescribing additional medication potentially do not outweigh the risks. ³ Prescribing cascades can lead to excess medication burden, which is of particular concern in older adults. Nonetheless, prescribing cascades can occur at any age and are more likely to occur in people using multiple medications. ⁴ Multimorbidity and polypharmacy as well as aging may complicate the recognition of ADRs, making it difficult to identify prescribing cascades. ³ , ⁵ , ⁶

Previous studies have identified and assessed prescribing cascades but few aimed to summarize those relevant for clinical practice. ⁵ , ⁷ , ⁸ , ⁹ , ¹⁰ , ¹¹ , ¹² , ¹³ A scoping review identified 281 signals for potential prescribing cascades based on medication prescription sequences. ⁵ This included potential prescribing cascades for which there was no prior evidence that the condition treated was an ADR, which limits their clinical relevance. Another review identified 94 prescribing cascades, including some that might be considered appropriate and others that lacked evidence for the potential ADR. ⁸ In another study, an expert panel assessed 139 potential prescribing cascades and classified nine of these as clinically important. ¹⁰ Although this assessment could include a consideration of how common the prescribing cascade might be, no quantification was provided to confirm its occurrence in clinical practice. Most studies that quantify prescribing cascades include only one prescribing cascade. ¹¹ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ In one study, 12 potential prescribing cascades were included—involving cardiovascular medication—and nine of these could be confirmed in a national database. ¹² Two studies investigated the concurrent use of multiple medications potentially causing the same ADR and observed that this increased the risk of a prescribing cascade. ²⁰ , ²¹

To better address prescribing cascades and decrease related medication burden in older adults, having an overview of prescribing cascades that are potentially problematic and occur in clinical practice is important. Only prescribing cascades that can be recognized by healthcare providers should be included. For this, there must be evidence that the ADR involved can be caused by the initial medication. Given the notion that inappropriate prescribing can start before adults are considered older adults, ⁴ , ²² the overview may well include prescribing cascades occurring regardless of age. Therefore, the primary aim of this study is to identify and quantify potentially problematic prescribing cascades among adults. A secondary aim is to explore the influence of concurrent use of multiple medications, potentially causing the same ADR, on the occurrence of prescribing cascades.

METHODS

Study design

A mixed‐methods study was performed. First, a literature search was conducted to generate an overview of prescribing cascades relevant for clinical practice. These prescribing cascades were subsequently assessed by experts to identify potentially problematic prescribing cascades. Next, a cohort study was performed using a Dutch community pharmacy database to quantify the occurrence of these potentially problematic prescribing cascades using prescription sequence symmetry analysis (PSSA). Additional analyses were conducted to assess the impact of multiple medications.

Ethics

The Advice Committee Scientific Research (ACWO) of OLVG Hospital approved this study (WO20132). Only anonymized patient data were used.

Part one

Literature search

A search was performed in Pubmed in July 2020 to identify prescribing cascades described in observational studies. The following search strategy was developed based on an earlier review: ⁴ (“prescribing cascade*”[tiab] OR “prescription cascade*”[tiab] OR “prescribing sequence*”[tiab] OR “prescription sequence*”[tiab]). Publications were evaluated by two researchers on eligibility. Included were publications in English that reported on one or more prescribing cascades observed in clinical practice. Publications were excluded when the prescribing cascades were not studied in adults (≥18 years), when prescribing cascades were not described in an observational study, or the study involved discovery‐driven analyses. The latter concerns, for example, pharmacovigilance studies using data‐mining techniques to detect new ADRs by analyzing prescription sequence signals. Finally, reviews were excluded. Snowballing was performed by checking the references of reviews and the references of the included publications.

Inclusion of prescribing cascades

A prescribing cascade was included when the initial medication (index medication), the potential ADR, and the subsequently prescribed medication (marker medication) were described. To identify prescribing cascades relevant for clinical practice, the following exclusion criteria were applied: (i) index medication has <20,000 users in a national database to exclude an index medication that is used by ≤0.1% of the Dutch population; ²³ (ii) marker medication is not indicated for treatment of the ADR in the Netherlands; (iii) ADR cannot be confirmed in Micromedex® ²⁴ for the index medication; (iv) index medication is commonly intended for short‐term use, that is, a treatment duration of <1 month for the main indication in the Netherlands; (v) index medication is described at an unspecific level, for example, the highest level in the Anatomical Therapeutic Chemical (ATC) classification system; ²⁵ and (vi) treatment of the ADR with the marker medication is recommended, as derived from guidelines or stated in the publication (e.g., prescribing a laxative to prevent constipation of opioids). ²⁶

Grouping of prescribing cascades

Similar prescribing cascades were grouped at medication and ADR level to form distinct prescribing cascades (Data S1). The ATC classification system ²⁵ was used to group medications that were described both at substance level (e.g., metoprolol) and at the related subgroup level (e.g., beta‐blocking agents). When the ADR could not be confirmed in Micromedex® for at least half of the substances within a subgroup, the prescribing cascade was presented at substance level. Similar ADRs (e.g., dribbling of urine and urinary incontinence) were grouped using the Medical Dictionary for Regulatory Activities classification system. ²⁷ Senior researchers (FKC, JH) verified the grouping of distinct prescribing cascades.

Assessment of prescribing cascades

An expert panel assessed the distinct prescribing cascades through an online questionnaire using Survalyzer® (Survalyzer Nederland BV, Utrecht, the Netherlands). The experts were general practitioners, internal medicine specialists, geriatricians, hospital pharmacists, community pharmacists, and an ADR expert from the Dutch pharmacovigilance center LAREB. For each prescribing cascade, they answered the question “to what extent do you consider this prescribing cascade to be problematic?” using a Visual Analogue Scale ranging from 0 to 100. They were given the following definition of problematic: “the benefits of a prescribing cascade do not outweigh the risks for the health of a patient.” Experts could provide additional problematic prescribing cascades encountered in practice. The Visual Analogue Scale was divided into three sections, not problematic (0–45), neutral (46–54), and problematic (55–100). ²⁸ Agreement was reached if ≥70% of the answers were in the same section. ²⁸ Prescribing cascades without agreement and those added by experts were discussed in an online meeting to reach consensus, resulting in a final overview of potentially problematic prescribing cascades.

Part two

Study design

Potentially problematic prescribing cascades were quantified with PSSA in a cohort study using community pharmacy dispensing data. The Joanna Briggs Institute Critical Appraisal checklist for cohort studies was followed. ²⁹

Setting

Medication data were collected from the Ncontrol database. ³⁰ Ncontrol includes data from community pharmacies located in urban and rural areas across the Netherlands. Data from January 1, 2015 to December 31, 2020 were retrieved for >600 affiliated pharmacies with complete records for that period, covering 30% of all Dutch community pharmacies. Patients in the Netherlands typically register at one pharmacy, where they collect all outpatient prescriptions. Prescription data are commonly transferred when a patient switches between pharmacies to obtain an adequate medication history. Over‐the‐counter purchases in the pharmacy can be included but purchases elsewhere are not captured in the database.

Study population included according to prescription sequence symmetry analysis parameters

Adult patients (≥18 years) were included who were dispensed both the index and marker medication of a prescribing cascade and complied with four key PSSA parameters as specified in a previous study. ³¹ First, a 12‐month washout window was applied, defined as a medication‐free period to ensure the index or marker medication is the first dispensing (incident user). ³¹ Second, a 12‐month exposure window was applied, requiring the index and marker medication to be dispensed within this period. ³¹ This reduces confounding factors (e.g., aging and disease progression). Third, a 4‐month continued exposure interval was applied to increase the likelihood of simultaneous use of index and marker medication. This interval is the gap between the expected end date of an index medication and the dispensing date of a marker medication (and vice versa), ³¹ , ³² establishing a maximum acceptable period without medication supply and accounting for any medication the patient may have in stock. ³³ Finally, a 7‐day blackout period was applied between the dispensing dates of the index and marker medication. This period ensures sufficient time for the development and reporting of the ADR, reducing the chance or an unrelated initiation of the marker medication. ³¹ , ³⁴

Data collection and classification

Data extraction was based on the ATC‐codes of the medication. The index medication was extracted at the pharmacological subgroup level, except for lithium and amiodarone. The latter were extracted at substance level due to the differences in mechanism of action and associated ADRs at subgroup level. For the marker medication, all subgroups related to the potential ADR were extracted (e.g., in case of a prescribing cascade involving hypertension as potential ADR treated with an angiotensin‐converting enzyme (ACE)‐inhibitor, also other antihypertensives were extracted). The extracted data included an anonymous patient identifier, year of birth, sex, dispensing dates, and expected end dates for prescriptions of the index and marker medication.

For the secondary aim, prescribing cascades with a significant positive association between the index and marker medication (see data analysis) were grouped on similar ADR‐marker medication. For each ADR‐marker group, patients were classified into: (1) patients using only one index medication (single users) before the marker medication or vice versa, (2) patients using two index medications and—when occurring—patients using three or more index medications (multiple medication users) before the marker medication or vice versa (Figure 1). When a patient used an index medication before the marker medication and a different index medication after the marker medication, this patient was included as a single user for the first index medication. Thus, only the first sequence over time was included. Since this method can bias toward higher aSRs due to not including some of the marker‐index sequences, a sensitivity analysis was performed excluding patients that received an index medication before the marker medication and a different index medication after the marker (see Figure S2). Combination products with multiple index medications (e.g., enalapril/hydrochlorothiazide) were counted at individual medication level.

FIGURE 1.

Examples of single users and multiple medications users for the adverse drug reaction‐marker group of medications potentially causing edema treated with the diuretic furosemide. The first sequence of incident medication use determines whether the patient complies with the sequence index followed by marker or vice versa. PSSA, prescription sequence symmetry analysis.

Outcome measures

The primary outcome was the association between the index and marker medication.

Data analysis

PSSA was performed to quantify the potentially problematic prescribing cascades. ³¹ The crude sequence ratio (cSR) was calculated as the ratio of patients receiving the index medication before the marker medication (sequence index‐marker [IM]) to those receiving the marker medication before the index medication (sequence marker‐index [MI]).

$$\text{crude sequence ratio}=\frac{\text{number of patients starting an index medication}\to \text{marker medication}\left(\mathrm{IM}\right)}{\text{number of patients starting}\mathrm{a}\text{marker medication}\to \text{index medication}\left(\mathrm{MI}\right)}$$

The cSR may be influenced by temporal prescribing trends. To account for this, the null‐effect sequence ratio (nSR) was calculated as an estimate of the overall probability for the index medication to be dispensed before the marker medication based on the prescription pattern in the background population for each prescribing cascade. ³⁵ , ³⁶ This overall probability is a running average for the patients included for each prescribing cascade for each day of the study period. ³¹ The background population exists of patients using both index and marker medication irrespective of the sequence. The adjusted sequence ratio (aSR) was calculated by dividing the cSR by the nSR.

$$\text{adjusted sequence ratio}=\frac{\text{crude sequence ratio}}{\text{null}-\text{effect sequence ratio}}$$

An aSR with a lower limit of the 95% confidence interval [95% CI] >1.0 represents a significant positive association between the index and marker medication, supporting the occurrence of a prescribing cascade. When comparing the aSR of single users with that of multiple medications users, nonoverlapping confidence intervals were considered to show a difference between the groups. A post hoc analysis was conducted on potentially problematic prescribing cascades with a statistically significant aSR, to quantify these among older adults (≥65 years).

To validate the Ncontrol database, the association between amiodarone and allopurinol was calculated as a negative control, as was done in previous studies. ¹² , ³⁷ , ³⁸ Amiodarone does not cause gout and thus an aSR around 1.0 is expected.

All analyses were performed using IBM SPSS version 29 (IBM Corporation, Armonk, New York, United States).

RESULTS

The literature search resulted in 147 publications. Of these, 81 were excluded because they did not study prescribing cascades (n = 50), did not describe prescribing cascades in an observational study (n = 15), did not study prescribing cascades in adults (n = 5), were data‐mining studies (n = 6), or were reviews (n = 5). In the remaining 66 publications, 303 prescribing cascades were identified of which 167 were excluded. The main reason for exclusion was that the index medication had <20,000 users in the Netherlands (Figure 2). The remaining 136 prescribing cascades were grouped into 76 distinct prescribing cascades.

FIGURE 2.

Flowchart of the evaluated prescribing cascades from 147 identified publications. ADR, adverse drug reaction; ATC, anatomical therapeutic chemical.

Assessment of prescribing cascades

The 76 distinct prescribing cascades were assessed by experts (Table S3). Sixteen of 19 approached experts completed the online questionnaire: four general practitioners, three community pharmacists, three hospital pharmacists, three geriatricians, two internal medicine specialists, and one ADR expert. Of 76 prescribing cascades, 62 (81.6%) were considered problematic by agreement based on the online questionnaire. The remaining 14 prescribing cascades were discussed in an online meeting with 11 experts (three general practitioners, two community pharmacists, two hospital pharmacists, two geriatricians, one internal medicine specialist, and one ADR expert). For 13 prescribing cascades either no consensus could be reached or consensus was reached that they were not problematic. For one prescribing cascade, consensus was reached that it was problematic. The experts provided three additional prescribing cascades that were all considered to be problematic (Table S3). This resulted in a total of 66 prescribing cascades to be quantified with PSSA.

Quantifying prescribing cascades

The negative control amiodarone with allopurinol showed an aSR of 0.91 [95% CI 0.65–1.18], supporting the validity of the PSSA method. Of 66 potentially problematic prescribing cascades, 41 had a significant aSR >1.0 (Table 1; Table S4). Frequently involved index medications were ACE‐inhibitors, antidepressants, lithium, and 3‐hydroxy‐3‐methyl‐glutaryl‐coenzyme A reductase‐inhibitors (statins). Depression and erectile dysfunction were the most frequently occurring potential ADRs. The highest aSR was for amiodarone potentially causing hypothyroidism treated with thyroid hormones (aSR 4.63 [95% CI 4.40–4.85], based on 565 incident users). The prescribing cascade with the largest number of incident users was ACE‐inhibitors potentially causing urinary tract infections treated with antibiotics. Ten prescribing cascades had an aSR ≥2.0, including lithium potentially causing tremor treated with propranolol and antipsychotics potentially causing hyperprolactinemia treated with prolactin inhibitors (Table 1). For amiodarone potentially causing hyperthyroidism treated with carbimazole no patients were included after applying the inclusion criteria based on the PSSA parameters. The remaining 25 prescribing cascades showed a nonsignificant or inverse association between the index and marker medication (Table S5).

TABLE 1.

Prescribing cascades with a significant association between index and marker medication, ranked from the highest to the lowest adjusted sequence ratio.

	Prescribing cascades
	Index medication	Potential ADR	Marker medication	Mean age [SD]	Female (%)	Incident users [IM/MI a ]	cSR	aSR [95% CI]
1	Amiodarone	Hypothyroidism	Thyroid hormones	78 [10]	52.0	565 [476/89]	5.35	4.63 [4.40–4.85]
2	Lithium	Tremor	Propranolol	54 [15]	63.3	150 [114/36]	3.17	2.91 [2.54–3.29]
3	ACE‐inhibitors	Cough	Cough and cold preparations	68 [13]	54.1	20,313 [15,121/5192]	2.91	2.59 [2.56–2.62]
4	HMG CoA reductase inhibitors	Cognitive impairment	Anti‐dementia medications	79 [7]	47.6	691 [497/194]	2.56	2.56 [2.20–2.53]
5	Proton pump inhibitors	Clostridium difficile infection	Intestinal antiinfectives	62 [16]	66.4	8281 [6068/2213]	2.74	2.49 [2.44–2.54]
6	Lithium	Parkinsonism	Tertiary amines/Dopaminergics	55 [16]	56.8	95 [66/29]	2.28	2.18 [1.74–2.62]
7	Lithium	Hypothyroidism	Thyroid hormones	55 [15]	81.1	419 [294/125]	2.35	2.17 [1.96–2.38]
8	Antipsychotics	Hyperprolactinemia or Oligomenorrhea	Prolactin inhibitors	40 [11]	74.2	31 [23/8]	2.88	2.12 [1.32–2.93]
9	Antipsychotics	Parkinsonism	Tertiary amines/Dopaminergics	55 [21]	46.3	2248 [1543/705]	2.19	2.08 [1.99–2.16]
10	ACE‐inhibitors	Cough	Antibacterials (systemic use)	68 [14]	48.5	12,313 [8557/3756]	2.28	2.04 [2.00–2.08]
11	ACE‐inhibitors	Urinary tract infections	Antibacterials (systemic use)	69 [14]	51.8	34,645 [23,424/11,221]	2.09	1.91 [1.89–1.94]
12	ACE‐inhibitors	Erectile dysfunction	Medications used in erectile dysfunction	65 [10]	0.5	3188 [2152/1036]	2.08	1.91 [1.84–1.99]
13	HMG CoA reductase inhibitors	Erectile dysfunction	Medications used in erectile dysfunction	65 [10]	0.4	5003 [3322/1681]	1.98	1.82 [1.77–1.88]
14	Dihydropyridines	Edema peripheral	High‐ceiling diuretics	76 [12]	56.3	10,317 [6800/3517]	1.93	1.82 [1.77–1.86]
15	Angiotensin II receptor blockers	Erectile dysfunction	Medications used in erectile dysfunction	66 [10]	0.6	1961 [1289/672]	1.92	1.80 [1.70–1.89]
16	Dihydropyridines	Erectile dysfunction	Medications used in erectile dysfunction	66 [9]	0.5	2791 [1825/966]	1.89	1.76 [1.68–1.84]
17	ACE‐inhibitors	Cough	Antihistamines (systemic use)	64 [14]	53.3	12,191 [7804/4387]	1.78	1.68 [1.64–1.71]
18	Beta blocking agents	Erectile dysfunction	Medications used in erectile dysfunction	66 [10]	0.8	3035 [1937/1098]	1.76	1.66 [1.58–1.73]
19	Low‐ceiling diuretics	Erectile dysfunction	Medications used in erectile dysfunction	66 [9]	0.3	1980 [1251/729]	1.72	1.61 [1.52–1.70]
20	Non‐dihydropyridines	Erectile dysfunction	Medications used in erectile dysfunction	67 [9]	1.8	282 [177/105]	1.69	1.60 [1.36–1.84]
21	HMG CoA reductase inhibitors	Arrhythmia	Antithrombotics	74 [11]	39.3	15,437 [9285/6152]	1.51	1.47 [1.43–1.50]
22	HMG CoA reductase inhibitors	Sleeplessness	Hypnotics and sedatives	68 [13]	54.7	15,358 [9280/6078]	1.53	1.45 [1.41–1.48]
23	HMG CoA reductase inhibitors	Urinary incontinence	Medications for urinary frequency and incontinence	72 [10]	17.5	12,025 [7202/4823]	1.49	1.42 [1.38–1.46]
24	Low‐ceiling diuretics	Gout	Anti‐gout medication	69 [12]	29.0	2032 [1201/831]	1.45	1.39 [1.30–1.47]
25	High‐ceiling diuretics	Erectile dysfunction	Medications used in erectile dysfunction	69 [12]	8.9	688 [405/283]	1.43	1.38 [1.23–1.53]
26	Antiepileptics	Urinary tract infections	Antibacterials (systemic use)	64 [17]	64.2	15,332 [9006/6326]	1.42	1.37 [1.34–1.41]
27	ACE‐inhibitors	Arthritis	Antiinflammatory & antirheumatic medication	64 [13]	47.0	24,650 [14,615/10,035]	1.46	1.37 [1.34–1.40]
28	HMG CoA reductase inhibitors	Agitation	Antipsychotics/Benzodiazepines	67 [13]	54.8	26,267 [15,414/10,853]	1.42	1.35 [1.33–1.38]
29	Antidepressants	Migraine	Analgesics and antipyretics	65 [19]	70.4	19,276 [11,125/8151]	1.36	1.33 [1.30–1.35]
30	HMG CoA reductase inhibitors	Confusion state	Antipsychotics	65 [15]	48.8	4600 [2633/1967]	1.34	1.28 [1.23–1.34]
31	ACE‐inhibitors	Cough	Adrenergics, inhalants	68 [13]	51.2	15,884 [8894/6990]	1.27	1.23 [1.20–1.27]
32	HMG CoA reductase inhibitors	Depression	N‐S MRI/ SSRI/antidepressants	65 [13]	54.9	17,106 [9407/7699]	1.22	1.18 [1.15–1.21]
33	Non‐dihydropyridines	Oedema peripheral	High‐ceiling diuretics	75 [12]	60.6	1605 [877/728]	1.20	1.18 [1.08–1.28]
34	Antidepressants	Parkinsonism	Tertiary amines/ Dopaminergics	62 [18]	59.0	3058 [1657/1401]	1.18	1.16 [1.09–1.23]
35	Beta blocking agents	Depression	N‐S MRI/ SSRI/antidepressants	62 [18]	63.9	17,658 [9547/8111]	1.18	1.15 [1.12–1.18]
36	Antiepileptics	Oedema peripheral	High‐ceiling diuretics	74 [13]	58.5	4698 [2528/2170]	1.16	1.15 [1.10–1.21]
37	Diuretics	Diuresis excessive	Urologicals	75 [12]	57.6	3296 [1789/1507]	1.19	1.15 [1.08–1.21]
38	ACE‐inhibitors	Depression	N‐S MRI/ SSRI/antidepressants	66 [14]	54.6	10,303 [5552/4751]	1.17	1.14 [1.10–1.18]
39	Antiinflammatory & antirheumatic medication	Oedema peripheral	High‐ceiling diuretics	73 [14]	63.9	11,363 [6023/5340]	1.13	1.13 [1.09–1.17]
40	Dihydropyridines	Depression	N‐S MRI/ SSRI/antidepressants	68 [14]	60.3	9046 [4791/4255]	1.13	1.11 [1.07–1.15]
41	Antidepressants	Urinary incontinence	Medications for urinary frequency and incontinence	66 [16]	35.4	6737 [3491/3246]	1.08	1.07 [1.02–1.12]

Abbreviations: ACE, angiotensin converting enzyme; ADR, adverse drug reaction; aSR, adjusted sequence ratio; CI, confidence interval; cSR, crude sequence ratio; high‐ceiling diuretics, loop diuretics; HMG CoA, 3‐hydroxy‐3‐methyl‐glutaryl‐coenzyme A; low‐ceiling diuretics, thiazide diuretics N‐S MRI, non‐selective monoamine reuptake inhibitors; SD, standard deviation; SSRI, selective serotonin reuptake inhibitors.

^a IM/MI = the number of patients who initiated the index medication before the marker medication (IM), the number of patients who initiated the marker medication before the index medication (MI).

In the post hoc analysis, 37 out of the 41 prescribing cascades also showed a significant positive aSR among older adults (Table S6). For four prescribing cascades with antipsychotics or antidepressants as index medications, the aSR became nonsignificant or could not be calculated.

Multiple index medications potentially causing the same adverse drug reaction

For seven ADR‐marker groups, more than one prescribing cascade with a significant positive aSR was found (Table S7). For one of these (i.e., hypothyroidism treated with thyroid hormones) there were no patients who used both index medications (i.e., lithium and amiodarone). The other ADR‐marker groups involved depression, parkinsonism, urinary incontinence, urinary tract infections, erectile dysfunction, and edema as potential ADR (Table S7). The use of multiple medications potentially causing these ADRs resulted in higher aSRs, except for the ADR edema potentially caused by ≥3 medications (Figures 3 and 4). In this latter case, only 49 patients were included resulting in a nonsignificant aSR. For four ADR‐marker groups the confidence intervals between single users and multiple medications users did not overlap (Figure 3). For two groups there was an overlap between single users and multiple medications users (Figure 4).

FIGURE 3.

Comparison of adjusted sequence ratios (aSRs) between single users and multiple medication users of index medications with number of included patients and their mean age for adverse drug reaction‐marker groups without overlap in 95% confidence intervals of the aSRs between these groups. ACE, angiotensin converting enzyme; aSR, adjusted sequence ratio; CI, confidence interval; HMG CoA, 3‐hydroxy‐3‐methyl‐glutaryl‐coenzyme A; MM, multiple medications; SD, standard deviation.

FIGURE 4.

Comparison of adjusted sequence ratios (aSRs) between single users and multiple medication users of index medications with number of included patients and their mean age for adverse drug reaction‐marker groups with overlap in 95% confidence intervals of the aSRs between these groups. ACE, angiotensin converting enzyme; aSR, adjusted sequence ratio; CI, confidence interval; high‐ceiling diuretics, loop diuretics; HMG CoA, 3‐hydroxy‐3‐methyl‐glutaryl‐coenzyme A; low‐ceiling diuretics, thiazide diuretics; SD, standard deviation.

When patients were excluded that had multiple sequences of the index and marker medication, similar results were found (Table S8). As expected, the aSRs for single users and also sometimes for multiple medications users became slightly lower but all differences between the groups remained.

DISCUSSION

We identified and quantified 66 potentially problematic prescribing cascades. A statistically significant positive association between the index and marker medication was observed for 41 (62.1%) of these prescribing cascades. These associations remained significant for 37 of 41 prescribing cascades when limiting the analysis to older adults. To assess the effect of multiple medications on the occurrence of a prescribing cascade, we identified seven ADR‐marker groups. Stronger associations between the index and marker medication were observed in multiple medications users compared to single users for potential ADRs of depression, parkinsonism, urinary incontinence, and urinary tract infections.

Comparisons between overviews of prescribing cascades are hampered because of differences in applied inclusion criteria. To date, none of the reviews have focused on older adults. McCarthy et al. reported nine clinically important prescribing cascades based on an assessment by experts. ¹⁰ In the present study, six of these were also labeled as problematic with three showing a significant positive association between the index and marker medication. The remaining three prescribing cascades had been suggested by experts in the McCarthy et al. study but were not identified from literature in the present study. Dreischulte et al. presented seven examples of prescribing cascades identified from literature, for which the marker medication was considered unsuitable or could be replaced by a safer alternative. ¹³ Five of these were also considered problematic in the present study but only three showed a significant positive association between the index and marker medication. These findings demonstrate the relevance of using observational studies with appropriate methodology to confirm the occurrence of prescribing cascades in clinical practice. ⁸ , ¹² As was stated by Shahid et al., 15 of the 94 prescribing cascades they evaluated could not be supported by high‐quality observational studies, ⁸ limiting their relevance for healthcare providers. In the present study, complying with quality criteria for cohort and PSSA studies, significant positive aSRs were observed for 62% of the tested prescribing cascades. This supports the occurrence of such a prescribing cascade at population level but it does not imply that each case concerns an actual prescribing cascade. Of note, at individual healthcare provider level some of these prescribing cascades may seldom or never occur. The strongest association was found for amiodarone potentially causing hypothyroidism treated with thyroid hormones both for the total study population and for older adults, based on limited numbers. The strength of the association between index and marker medication is influenced by the actual occurrence of the prescribing cascade but also by the timing of the potential ADR in relation to the time windows set for the PSSA. Various prescribing cascades with either weak or strong associations between the index and marker medication in the present study were confirmed with strong evidence from previous studies. ⁸ In the present study, we found a nonsignificant association for 25 prescribing cascades that had been rated as potentially problematic. This does not imply that these prescribing cascades never occur or are irrelevant for clinical practice. However, we could not detect them with our PSSA approach. This may have been due to uncommon occurrence in our study setting or period, or due to ADRs manifesting later than anticipated or being managed with a different marker medication than expected.

Despite being assessed as problematic in general, some prescribing cascades are unavoidable in individual cases that relate to a condition or situation for which there is no alternative treatment. Some experts found it difficult to assess whether a prescribing cascade is problematic in general, since they felt that the evaluation of benefits and risks depends on the patient context. Sometimes, healthcare providers are aware of an ADR and choose to manage it with medication. Nevertheless, such intentional prescribing cascades can still be problematic. Prescribing additional medication may not be the most appropriate action, particularly when this increases medication risks and burden in older adults. The finding that aSRs increased when people use multiple medications adds to previously reported findings. ²⁰ , ²¹ These studies considered multiple psychotropic medications that may cause parkinsonism, ²⁰ and multiple medications that may cause constipation. ²¹ However, the present study showed that prescribing cascades of antipsychotics and antidepressants that may cause parkinsonism appear uncommon among older adults. Healthcare providers may be more aware of the risks of such psychotropic medications in older adults.

Strengths and limitations

This study has several strengths. A stepwise approach was used to identify and select potentially problematic prescribing cascades relevant for clinical practice. This included a literature review and a systematic consensus procedure with a multidisciplinary group of experts. To quantify prescribing cascades, PSSA was conducted on a nationwide database and recommended parameters were applied to reduce bias. The approach was validated using a negative control.

There are also limitations. First, the literature search may not have identified all potential prescribing cascades, since terms like prescribing cascade or prescription sequence are not always used in publications. A systematic review conducted later using broader terms identified additional prescribing cascades, particularly concerning the ADRs dizziness and hyperglycaemia. ⁹ Second, prescribing cascades were excluded when the index medication was prescribed to ≤0.1% of the Dutch population or was commonly indicated for short‐term use. This may have resulted in excluding prescribing cascades that are relevant to other countries or settings. Third, confirmation of ADRs being linked to the index medication was based on data from the Micromedex® database. ²⁴ This database does not provide data on the quality of its information and can include ADRs with limited evidence. Fourth, although confounding is considered low in PSSA, there may be residual confounding that is insufficiently addressed by applying the recommended parameters with the risk of false positive findings. ³⁹ , ⁴⁰ Ideally, to confirm the potential ADR, information is needed on the indication of the marker medication but such information is not available in dispensing databases. When identifying single and multiple medications users, including only the first sequence involving the marker medication per patient introduces bias toward higher aSRs. However, our sensitivity analysis showed that this bias was small and appeared similar for single and multiple medications users. Nonetheless, applying PSSA for multiple medications as conducted has not been validated. When comparing the aSRs between single users and multiple medications users, no adjustment for confounding was made, whereas multiple medications users may be more prone to prescribing cascades due to multimorbidity. Finally, the end dates for prescriptions were estimated in the pharmacy information systems based on the provided dosing schedules. When medication is used as needed or patients are non‐adherent, the actual medication exposure differs from this calculated exposure, which may lead to both overestimations and underestimations of aSRs.

Implications

This study underlines that there is a substantial number of potentially problematic prescribing cascades that may occur in clinical practice, also among older adults. Healthcare providers should be aware of these potential prescribing cascades and address them where relevant and possible. Recently, it was suggested that more awareness of potential ADRs and/or prescribing cascades might explain why older adults ≥85 years were less likely to experience a prescribing cascade. ⁴¹ By preventing or reversing prescribing cascades in a timely manner, medication burden can be decreased for older adults. When conducting medication reviews in older adults with polypharmacy, healthcare providers can use the presented overview to screen for potential prescribing cascades. The overview could also be incorporated in tools and handbooks for clinical practice. ¹⁰ There is a need for research on healthcare technologies (e.g., decision support systems) making use of such information to prevent and manage prescribing cascades. Future research should assess the effects of interventions to prevent and reverse potentially problematic prescribing cascades.

The additive effect of multiple medications potentially causing the same ADR on the occurrence of prescribing cascades is of particular relevance for older adults, who often use multiple medications. When an older adult on medication presents with a new symptom, the first arising question should be: “could this be caused by medication?” Further research is needed to understand the impact of multiple medications on prescribing cascades and patient‐relevant outcomes, especially in older adults.

CONCLUSION

For 41 of 66 prescribing cascades identified as being potentially problematic, a significant positive association between the index and marker medication was observed, which indicates their occurrence in clinical practice. Most could be confirmed among older adults. Furthermore, seven ADR‐marker groups were identified for prescribing cascades linked to multiple medications. The use of multiple medications generally increased the likelihood of a prescribing cascade. This overview can support healthcare providers in addressing potentially problematic prescribing cascades.

AUTHOR CONTRIBUTIONS

Atiya K. Mohammad contributed to the acquisition of data, data analysis, interpretation of data, and drafting and revision of the manuscript. Jacqueline G. Hugtenburg, Patricia M. L. A. van den Bemt, and Petra Denig contributed to the study concept and design, interpretation of data, and critical review of manuscript. Joost W. Vanhommerig contributed to the data analysis, interpretation of data, drafting and revision of the manuscript, and critical review of the manuscript. Fatma Karapinar‐Carkıt contributed to the study concept and design, interpretation, the acquisition of data, data analysis, interpretation of data, drafting and revision of the manuscript, and critical review of manuscript.

CONFLICT OF INTEREST STATEMENT

The Royal Dutch Association of Pharmacists (KNMP) has supported the study with a non‐conditional grant (Grant number PR20_0103). The funders had no role in the study design, data collection, data analysis, decision to publish, or preparation of the manuscript. The authors declare that none of them have received honoraria, reimbursement, or fees from any pharmaceutical companies, related to this study.

SPONSOR'S ROLE

Not applicable.

FINANCIAL DISCLOSURE

The Royal Dutch Association of Pharmacists (KNMP) has supported the study with a non‐conditional grant (No. PR20_0103).

Supporting information

Data S1. Grouping of similar prescribing cascades for the expert panel evaluation.

Figure S2. Examples of excluded patients for sensitivity analysis regarding multiple medications use.

Table S3. Prescribing cascades as assessed by experts.

Table S4. Details of prescribing cascades with significant associations between the index and marker medication.

Table S5. Prescribing cascades with nonsignificant or inverse significant associations between the index and marker medication or prescribing cascades with lack of data.

Table S6. Post hoc analyses among older adults (≥65 years of age) for prescribing cascades that showed a significant association between index and marker medication in adult patients.

Table S7. Details of single and multiple medications users for potential adverse drug reaction‐marker groups.

Table S8. Details of single and multiple medications users for potential adverse drug reaction‐marker groups in the sensitivity analysis excluding patients with different index medications, both before and after the marker medication.

Mohammad AK, Hugtenburg JG, Vanhommerig JW, van den Bemt PMLA, Denig P, Karapinar‐Carkıt F. Identifying and quantifying potentially problematic prescribing cascades in clinical practice: A mixed‐methods study. J Am Geriatr Soc. 2024;72(12):3681‐3694. doi: 10.1111/jgs.19191