Associations of Polypharmacy and Inappropriate Medications with Adverse Outcomes in Older Adults with Cancer: A Systematic Review and Meta‐Analysis

Abstract

Background

Polypharmacy (PP) and potentially inappropriate medications (PIM) are highly prevalent in older adults with cancer. This study systematically reviews the associations of PP and/or PIM with outcomes and, through a meta‐analysis, obtains estimates of postoperative outcomes associated with PP in this population.

Materials and Methods

We searched PubMed, Embase, Web of Science, and Cochrane Register of Clinical Trials using standardized terms for concepts of PP, PIM, and cancer. Eligible studies included cohort studies, cross‐sectional studies, meta‐analyses, and clinical trials which examined outcomes associated with PP and/or PIM and included older adults with cancer. A random effects model included studies in which definitions of PP were consistent to examine the association of PP with postoperative complications.

Results

Forty‐seven articles met the inclusion criteria. PP was defined as five or more medications in 57% of the studies. Commonly examined outcomes included chemotherapy toxicities, postoperative complications, functional decline, hospitalization, and overall survival. PP was associated with chemotherapy toxicities (4/9 studies), falls (3/3 studies), functional decline (3/3 studies), and overall survival (2/11 studies). A meta‐analysis of four studies indicated an association between PP (≥5 medications) and postoperative complications (overall odds ratio, 1.3; 95% confidence interval [1.3–2.8]). PIM was associated with adverse outcomes in 3 of 11 studies.

Conclusion

PP is associated with postoperative complications, chemotherapy toxicities, and physical and functional decline. Only three studies showed an association between PIM and outcomes. However, because of inconsistent definitions, heterogeneous populations, and variable study designs, these associations should be further investigated in prospective studies.

Implications for Practice

Polypharmacy and potentially inappropriate medications (PIM) are prevalent in older adults with cancer. This systematic review summarizes the associations of polypharmacy and PIM with health outcomes in older patients with cancer. Polypharmacy and PIM have been associated with postoperative complications, frailty, falls, medication nonadherence, chemotherapy toxicity, and mortality. These findings emphasize the prognostic importance of careful medication review and identification of PIM by oncology teams. They also underscore the need to develop and test interventions to address polypharmacy and PIM in older patients with cancer, with the goal of improving outcomes in these patients.

Short abstract

This review evaluates the association of polypharmacy and potentially inappropriate medications with outcomes in older adults with cancer.

Introduction

In the U.S., more than half of new cancer cases and about 70% of cancer deaths occur in adults aged 65 years or older 1. Compared with younger cohorts, older adults are more likely to have comorbid conditions for which medications are prescribed 2. Older adults with cancer also have a higher rate of frailty and geriatric syndromes compared with those without cancer 3. For patients receiving cancer treatment, chemotherapy and supportive care regimens often involve the prescription of multiple medications. Because of these factors, older adults with cancer are at high risk of polypharmacy (PP), defined as the simultaneous use of multiple medications. In community‐dwelling populations of older adults without cancer, PP has been associated with increased falls 4, hospitalization 5, and mortality 6.

There is a wide variability in the definition of PP in the existing literature 7. In one study of community‐dwelling older adults with cancer, 84% were on five or more medications and 43% were on 10 or more medications 8. The use of five or more medications is the most commonly used definition of PP in the literature, whereas the use of 10 or more medications is commonly referred to as “extreme PP” or “hyperpolypharmacy.” Multiple cutoffs are used in the literature, and studies vary as to how medications are counted (i.e., whether only scheduled prescription medications are included, or whether supplements, over‐the‐counter medications, and as‐needed medications are counted as well 8).

Definitions of PP typically do not account for the appropriateness of medications. PP increases the risk that one or more medications is “potentially inappropriate”; these potentially inappropriate medications (PIMs) have risks higher than anticipated benefits in older adults. PIMs may be assessed using multiple validated instruments including the Beers criteria 9, Screening Tool of Older Person's Prescriptions (STOPP) and Screening Tool to Alert to Right Treatment (START) criteria 10, Zhan criteria 11, and medication appropriateness index (MAI) 12. In older adults, PIMs are associated with increased risks of adverse drug events, hospitalizations, and mortality 13, as well as higher health care costs 14. However, supportive care regimens may include medications (such as benzodiazepines for treatment of nausea) that would otherwise be deemed PIM but may be clinically appropriate based on oncology supportive care guidelines.

Although the literature on PP and PIM in older adults is increasing, data in older adults with cancer remain sparse. Extrapolation of data from the general population of older adults is problematic: older adults with cancer have more frailty and multimorbidity than patients without cancer, they take more medications on average, and the initiation of chemotherapy and supportive care regimens can significantly increase the risk of drug‐drug interactions and adverse drug events 15. It also remains unclear whether PP and PIM affect outcomes in older adults with cancer. This systematic review and meta‐analysis evaluates the association of PP and PIM with outcomes in older adults with cancer.

Materials and Methods

Search Strategy

We followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta‐Analyses) guidelines 16. We searched for articles from the following databases between the database inception and September 2018: PubMed, Web of Science, Embase, and Cochrane Register of Controlled Trials. Standardized terms and keywords were combined in the search for the following concepts: oncology/cancer, polypharmacy, and inappropriate medications (supplemental online Appendix 1). Reference lists of relevant articles were screened to identify other relevant articles (“snowball” search). All results were exported to EndNote, and duplicates were identified and removed.

Selection Criteria

We included studies if they (a) examined any outcomes associated with PP and/or PIM; (b) included patients with cancer (as either the whole sample or a subgroup); (c) included adults aged ≥65 (a common cutoff to identify older adults in the literature); (d) were clinical trials, observational cohort studies, cross‐sectional studies, or meta‐analyses; and (e) were written in English. We excluded studies that did not specifically evaluate the associations of PP and/or PIM with outcomes (i.e., studies that described the prevalence of PP and/or PIM only) and those that were published in abstract form only. Two investigators (M.M. and A.A.) independently reviewed the titles and abstracts of retrieved articles to select potential articles. The full texts were further reviewed independently by M.M. and A.A. for final selection of articles. Disagreements were resolved by a third investigator (K.P.L.).

Data Extraction and Analysis

A predefined data extraction template was developed and included name of the first author, year of publication, country, study design, sample size, age, definition of PP and PIM, prevalence of PP and PIM, cancer type, treatment planned or received, outcome variables, and findings on the association between PP and PIM and outcome measures. Two independent authors (M.M. and A.A., K.P.L., or S.O.) reviewed full texts of each identified article and extracted the data.

After data extraction, outcomes were categorized into the following domains: postoperative outcomes, chemotherapy outcomes, physical function, survival, and miscellaneous/other outcomes. Odds ratios (OR), p values, and 95% confidence intervals (CI) were reported for significant results (defined as p < .05 or 95% CI did not cross 1) if available. Nonsignificant results are reported as NS, but p values and OR are not reported as most studies did not report these for nonsignificant results.

Meta‐Analysis

Studies evaluating the association of PP with postoperative complications used a consistent definition of PP (≥5 medications) and postoperative complications (using the Clavien‐Dindo classification; n = 4 studies) 17. We performed a random‐effects model to combine the OR and 95% CI in these four studies. Heterogeneity of included studies was measured using chi‐square test and the I² statistic, with a significant heterogeneity defined as I² > 50%. Forest plots present individual and pooled risk estimates. All statistical analyses were conducted using Stata 13.0 (Stata Corp, College Station, TX). Meta‐analyses for other outcomes were not undertaken because of heterogeneity of definitions of PP and PIM.

Quality Appraisal

Two independent authors (M.M. and A.A., K.P.L., E.R., or S.O.) assessed the quality of each selected study. We used the National Institutes of Health (NIH) Quality Assessment Tool for Observational Cohort and Cross‐Sectional Studies 18 as a guide. We rated the data quality related to PP and PIM and outcomes as good, fair, or poor. Disagreements among the reviewers were discussed and resolved during consensus meetings.

Results

Study Characteristics

The initial search strategy identified 3,459 titles and abstracts. An additional six articles were identified from the reference lists of selected articles (Fig. 1). In total, 47 studies were included (number of patients ranged from 16 to 40,009). These studies were published between 2005 and 2018 from 19 countries. Study designs included retrospective cohort (23 studies), prospective cohort (14 studies), cross‐sectional (9 studies), and meta‐analysis (1 study). Of these studies, 46.8% (22/47) included patients with only one cancer type. Common cancer types were gastrointestinal (GI; 24/47, 51.1%), breast (20/47, 42.5%), and lung (8/47, 17%).

Figure 1.

PRISMA flow diagram. This diagram details our search and study selection process applied during the study according to PRISMA checklist.

Quality Appraisal

The articles were judged to be good (n = 2, 4.2%), fair (n = 32, 68.1%), or poor quality (n = 13, 27.6%) based on the NIH Quality Assessment Tool (supplemental online Tables 1 and 2). All studies clearly stated their research objectives and study population. Sample sizes were clearly stated in all studies; however, only two studies provided a justification for the sample size. Only five studies assessed the exposure more than once. Based on the authors’ judgment, 19 of the 47 articles did not adequately adjust for potential confounding variables.

Definition and Prevalence of PP and PIM

When defining PP, only 19 studies clearly stated which types of medications were included in their analyses (i.e., prescription, supplemental, and/or over‐the‐counter medications; supplemental online Table 1). Overall, the prevalence of PP ranged from 2.0% to 80.0%. PP was defined as the use of five or more medications in 57.4% (27/47) of the studies; in these studies, the prevalence ranged from 14.0% to 80.0%. Other definitions included (a) 10 or more medications (7/47, with prevalence ranging from 5.6% to 43.0%), (b) 9 or more medications (1/47, with no prevalence reported), (c) 6 or more medications (3/47, 8.0% to 38.0%), (d) 4 or more medications (4/47, 49.0% to 86.0%), (e) 3 or more medications (2/47, 43.0% to 52.0%), (f) any concomitant (≥1) medications in addition to cancer treatment (2/47, 2.0%, and 69.0%), and (g) other definitions (2/47).

Overall, the prevalence of PIM ranged from 19.0% to 52.0% (assessed in 11/47 studies). Ten of 11 studies used the Beers criteria to screen for PIM. Several tools were used in addition to the Beers criteria including Drugs to Avoid in Elderly (DAE) list (2/47), STOPP criteria (1/47), HEDIS (1/47), National Board of Health and Welfare Criteria (1/47), and Zhan (1/47).

Outcomes

Among all studies, 77.0% (36/47), 6.0% (3/47), and 17.0% (8/47) investigated outcomes associated with PP, PIM, or both PP and PIM, respectively. Postoperative outcomes (e.g., postoperative complications, delirium, extended hospital stay, and emergency department visit after cancer surgery) were evaluated in 11/47 studies (23%; Table 1) 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29. Chemotherapy‐related outcomes, such as chemotherapy‐related toxicities, chemotherapy completion, hospitalization after chemotherapy, complete remission (CR), chemotherapy dose reductions or delay, and blood transfusion, were evaluated in 12/47 studies (26%; Table 2) 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41. Frailty, falls, and physical and functional outcomes were evaluated in 7/47 studies (15%; Table 3) 42, 43, 44, 45, 46, 47, 48. Survival outcomes were assessed in 12/47 studies (26%; Table 4) 22, 29, 30, 34, 36, 37, 49, 50, 51, 52, 53, 54. Other outcomes (e.g., medication adherence, caregiver burden and alternative medications use; supplemental online Table 3) 32, 39, 49, 53, 55, 56, 57, 58, 59, 60, 61, 62, 63.

Table 1.

Studies examined the association between PP and postoperative outcomes

Study	Study design	Sample size	Agea	Definition of PP	PP rate	Cancer type	Treatment	Outcome measures	Resultsb
McAlpine, 2008, Canada 24	Single center, retrospective cohort	103	Median: 72 (range, 60–91)	≥5 medications	Mean no. of medications preoperatively = 6.28, mean no. of medications postoperatively = 8.47	Gynecologic cancers	Surgery	Postoperative delirium	OR, 1.9; 95% CI, NR; p = .008
Kristjansson, 2010, Norway 22	Multicenter, prospective cohort	182	Median: 80 (range, 70–94)	≥5 medications	26.0%	Colorectal	Surgery	30‐d major postoperative complications (CD) and survival	NS
Badgwell, 2013, U.S. 20	Single center, prospective cohort	111	Median: 72 (range, 65–89)	≥5 medications	48.0%	GI	Intra‐abdominal surgery	Prolonged hospital stay, readmission within 30 days of surgery, and discharge to an SNF	Prolonged hospital stay: OR, 2.5; 95% CI, 1.1–5.5; p = .003 Other outcomes: NS
de Glas, 2013, The Netherlands 28	Multicenter, retrospective cohort	3,179	Median: 74 (range, 65–98)	≥5 medications	13.5%	Breast cancer	Breast cancer surgery	30‐d postoperative complications	OR, 1.8; 95% CI, 1.4–2.2; p < .001
Pujara, 2015, U.S. 23	Single center, retrospective cohort	279	Median: 64 (range, 25–88), 46% were ≥65 yrs	≥5 medications	20.1% (not specific to older adults)	Gastric	Intra‐abdominal surgery	Major postoperative complications (CD), prolonged hospital stay >14 d, and readmission within 30 d of surgery	OR, 2.4; 95% CI, 1.1–5.2; p = .03 Other outcomes: NS
Kenig, 2015, Poland 21	Single center, prospective cohort	75	Mean: 73 (SD 5.8; range, 65–93)	2 cutoff values were used: ≥4 or ≥ 5 medications	44.0%	GI	Intra‐abdominal surgery	30‐d major postoperative complications (CD)	≥4 medications: OR, 4.2; 95% CI, 1.4–12.1, p = NR ≥5 medications: OR, 2.8; 95% CI, 1.1–8.2, p = NR
Jeong, 2016, South Korea 27	Single center, retrospective cohort	475	Median: 76 (range, 65–96)	≥5 medications	50.5%	Any	Surgery	Postoperative delirium	NS
Westley, 2017, Canada 26	Population‐based retrospective cohort	24,463	Median: 74	PP: ≥5 medications EPP: ≥10 medications	31.8% and 5.6%	Breast cancer	Curative cancer surgery	Initial emergency department visit within 45 d of definitive breast cancer surgery	>5 medications: HR, 1.3; 95% CI, 1.2–1.3; p < .0001 ≥10 medications: HR, 1.5; 95% CI, 1.3–1.8; p < .0001
Fagard, 2017, Belgium 19	Single center, retrospective cohort	190	Median: 77 (61% 70–79, 39% >80)	≥5 medications	47.4%	Colorectal	Intra‐abdominal surgery	30‐d postoperative complications (CD)	NS
Choi, 2018, South Korea 25	Single center, retrospective cohort of prospective data	475	Median: 76 (range 65–96)	≥5 medications ≥10 medications	50.5% and 27.9%	Any	Cancer surgery	30‐d mortality and postdischarge institutionalization	Postdischarge institutionalization: OR, 3.8; 95% CI, 1.04–13.8; p < .05 Other outcome: NS

^aMedian or mean, SD (range if available).

^bMultivariate unless otherwise specified.

Abbreviations: CD, Clavien Dindo; CI, confidence interval; d, day; EPP, excessive polypharmacy; GI, gastrointestinal; HR, hazard ratio; NR, not reported; NS, nonsignificant; OR, odds ratio; PP, polypharmacy; SNF, skilled nursing facility.

Table 2.

Studies examined the association between PP and chemotherapy‐related outcomes

Study	Study design	Sample size	Agea	Definition of PP	PP rate	Cancer type	Treatment	Outcome measures	Resultsb
Sasaki, 2013, Japan 33	Single center, retrospective cohort	172 (71% ≥65)	Median: 64 (range, 31–78)	Any concomitant drug used to manage comorbid conditions besides cancer	69.0%	Solid tumors	Irinotecan‐based therapy	Severe toxicity (grade 4 neutropenia and/or grade 3/4 diarrhea)	OR, 4.7; 95% CI, 1.04–21.3; p = .04
Hamaker, 2013, The Netherlands 36	Single center, prospective study	78	Median: 76 (range, 66–87)	≥5 medications	51.0%	Breast cancer	First‐line chemotherapy	Grade 3–4 toxicity	OR, 6.4; 95% CI, 2.0–23.5; p = .001
Maggiore, 2014, U.S. 31	Multicenter, retrospective cohort	500	Median not reported: 54% ≥72	≥4 medications ≥10 medications	49.0% and 11.0%	Solid tumors	Outpatient chemotherapy	Chemotherapy‐related toxicity and hospitalization	NS
Elliot, 2014, U.S. 30	Single center, retrospective cohort	150	Median: 69 (range, 61–87)	≥5 medications ≥4 medications	≥5: 38.0%	Acute myeloid leukemia	Induction chemotherapy	CR status, prolonged hospital stay, and ICU admission	CR status: OR, 0.2; 95% CI, 0.1–0.6; p = NR Other outcomes: NS
Iurlo, 2014, Italy 35	Single center, retrospective cohort	16	Median: 72 (range, 65–88)	≥3 medications	43.7%	Chronic myeloid leukemia	Imatinib	CCR and change of chemotherapy regimen	CCR: 4/7 patients with PP vs. 5/9 patients without PPc Change of chemotherapy regimen: 2/7 patients with PP vs. 3/9 patients without PPc
Kim, 2014, South Korea 38	Single center, retrospective cohort	98	87.8% >70	≥6 medications	39.8%	Any cancer	Palliative chemotherapy	Early discontinuation of chemotherapy	NS
Ting, 2015, Singapore 39	Single center, cross‐sectional study	294	Mean: 71.8	≥5 medications	62.9%	Any cancer	Outpatient chemotherapy	Drug‐related problems (potential drug interactions, adverse events, and nonadherence)	OR, 3.2; 95% CI, 1.4–7.3; p = .006
Sud, 2015, Canada 40	Single center, retrospective cohort	318	Median: 82 (range, 80–92)	≥6 medications	38%	Solid tumors	Chemotherapy	Rate of discontinuation due to toxicity, rate of dose reduction/ omission/delay, hospitalization, and blood transfusion	Hospitalization: OR, 2.3; 95% CI, 1.3–3.9; p = .002 Other outcomes: NS
Park, 2016, Korea 32	Single center, retrospective cohort	229	Median: 73 (range, 65– 87)	≥5 medications	29.3%	Head and neck cancers	Surgery, radiotherapy, or chemo‐radiation	Grade 3–4 toxicity and hospitalization	NS
Iurlo, 2016, Italy 34	Multicenter, retrospective cohort	296	Mean: 79.4	≥5 medications	36.1%	Chronic myeloid leukemia	Imatinib	CCR, event‐free survival, and toxicities	NS
Woopen, 2016, Germany 37	Meta‐analysis (the original patient data of three phase II/III trials)	1,213 (% older adults not reported)	Median: 59 (range, 21–83)	≥5 medications	56.3%	Advanced ovarian cancer	Various chemotherapy regimens	Overall grade 3–4 toxicities, grade 3–4 hematological and nonhematological toxicity, and early discontinuation of chemotherapy	Overall grade 3–4 toxicities: OR, 1.1; 95% CI, NR; p < .001 Grade 3–4 hematological toxicities: OR, 1.056; 95% CI, NR; p < .001 Severe nonhematological toxicity: OR, 1.1; 95% CI, NR; p < .001 Other outcomes: NS
Antonio, 2018, Spain 41	Single center, prospective cohort	193	Mean: 79.6 (range, 75–89)	≥6 medications	64%	Colorectal cancer	Adjuvant chemotherapy with or without radiotherapy	Treatment refusal, grade 3 toxicity, and completion of at least 80% of the planned chemotherapy dose	Treatment refusal: OR, 5.3; 95% CI, 1.6–18.4; p = .01 Other outcomes: NS

^aMedian or mean, SD (range if available).

^bMultivariate unless otherwise specified.

^cNo analysis, only absolute numbers were reported.

Abbreviations: CI, confidence interval; CCR, complete cytogenic response; CR, completer response; HR, hazards ratio; ICU, intensive care unit; NR, not reported; NS, nonsignificant; OR, odds ratio; PP, polypharmacy.

Table 3.

Studies examined the association between PP and frailty, falls, and physical and functional outcomes

Study	Study design	Sample size	Agea	Definition of PP	PP rate	Cancer type	Treatment	Outcome measures	Resultsb
Prithviraj, 2012, U.S. 45	Single center, cross‐ sectional study	117	Mean: 74.6 (range, 65–85)	≥5 medications	80.0%	Solid tumors	Chemotherapy	IADL and VES score	IADL: OR, NR; 95% CI, NR; p = .007 (univariate only) VES score: OR, NR; 95% CI, NR; p = .03 (univariate only) PP was associated with IADL and VES score impairment
Vande Walle, 2014, Belgium 48	Multicenter, prospective cohort	937	Median: 76 (range, 70–95)	≥5 medications	53.1%	Any cancer	Not specified	Falls	NS
Turner, 2014, Australia 42	Single center, cross‐sectional study	385	Median: 76.7 (range, 70–92)	≥5 medications	57.0%	Any cancer	Not specified	Physical impairment, frailty, and prefrailty	Physical impairment: OR, 1.1; 95% CI, 1.1–1.2; p = NR Frail: OR, 4.5; 95% CI, 1.9–10.5; p = NR Prefrail: OR, 2.4; 95% CI, 1.4–3.8; p = NR
Williams, 2015, U.S. 47	Multicenter, cross‐sectional study	1,172	Mean: 73 (range, 65–99)	≥9 medications	Not reported	Any cancer	Not specified	Falls	OR, 1.6; 95% CI, 1.2–2.3; p < .001
Turner 2016, Australia 44	Single center, cross‐ sectional study	385	Median: 76.7 (range, 70–92)	≥5 medications	57.0%	Any cancer	Not specified	Falls, physical impairment, and frailty	≥5.5 medications was associated with increased fallsc ≥6.5 medications was associated with physical impairment and frailtyc
Pamoukdjian 2017, France 46	Multicenter, cross‐ sectional study	290	Mean: 80.6	≥5 medications	67.4%	Any cancer except HCC	Not specified	Disability (impairment in ADL and/or IADL)	OR, 2.3; 95% CI, 1.01–4.4; p = .04
van Abbema, 2017, The Netherlands 43	Multicenter, retrospective cohort	837 (21.4% ≥70)	Median: 60 (range, 50–69 y) Median in patients aged ≥70: 75 (range, 70–93)	≥5 medications	26.8% in patients aged ≥70	Breast and colorectal cancers	Surgery	Functional status (ADL, IADL, and combined)	ADL: OR, 2.1; 95% CI, 1.1–3.8; p = NR IADL: OR, 1.9; 95% CI, 1.1–3.4; p = NR Combined ADL and IADL: OR, 2.1; 95% CI, 1.3–3.5; p = NR

^aMedian or mean, SD (range if available).

^bMultivariate unless otherwise specified.

^cOdds ratio and p value were not reported (sensitivity and specificity values).

Abbreviations: CI, confidence interval; ADL, activity of daily living; IADL, instrumental activity of daily living; HCC, hepatocellular carcinoma; NR, not reported; NS, nonsignificant; OR, odds ratio; PP, polypharmacy; VES, Vulnerable Elders‐13 Survey.

Table 4.

Studies examined the association between PP and survival outcomes

Study	Study design	Sample size	Agea	Definition of PP	PP rate	Cancer type	Treatment	Outcome measures	Results
Freyer, 2005, France 52	Single center, prospective cohort	83	Median: 76	≥6 medications	8.0%	Ovarian cancer	First‐line palliative chemotherapy	40‐mo OS	PP was associated with lower OS (p = .04)b
Kristjansson, 2010, Norway 22	Multicenter, prospective cohort	182	Median: 80 (range, 70–94)	≥5 medications	26.0%	Colorectal	Surgery	30‐d major postoperative complication (CD) and survival	NS
Hamaker, 2013, The Netherlands 36	Single center, prospective study	78	Median: 76 (range, 66–87)	≥5 medications	51.0%	Breast cancer	First‐line chemotherapy	OS	NS
Falandry, 2013, France 51	Multicenter phase II clinical trial	111	Median: 79 (range, 71–93)	≥4 medications	68.0%	Ovarian cancer	First‐line carboplatin	OS	NS
Hamaker, 2014, The Netherlands 49	Single center, retrospective cohort	108	Median: 78.2 (range, 67.1–98.9)	≥5 medications	65.0%	Hematologic cancer	Standard treatment, adjusted treatment, or no treatment	1‐y mortality	NS
Elliot, 2014, U.S. 30	Single center retrospective cohort	150	Median: 69 (range, 61–87)	≥5 medications ≥4 medications	≥5: 38.0%	Acute myeloid leukemia	Induction chemotherapy	30‐d mortality and overall mortality	30‐day mortality: OR, 9.9; 95% CI, 1.2–84.1, p = NR Overall mortality: HR, 2.1; 95% CI, 1.2–3.9; p = NR
Woopen, 2016, Germany 37	Meta‐analysis the original patient data of three phase II/III trials	1,213	Median: 62	≥6 medications	56.0%	Ovarian cancer	Chemotherapy	Progression‐free survival and 5‐year OS	NS
Caparrotti, 2016, Canada 50	Single center, retrospective cohort	287	Median: 74–75	Comorbidity‐PP score (sum of comorbidity and medications), 6 or more	47.0%	Oropharyng‐eal cancer	Definitive radiation (chemotherapy is allowed)	5‐year OS	NS
Iurlo, 2016, Italy 34	Multicenter, retrospective cohort	296	Mean: 79.4	≥5 medications	36.0%	Chronic myeloid leukemia	Imatinib	OS (from date of imatinib initiation)	NS
Nieder, 2017, Norway 53	Single center, retrospective cohort	280	Median: 77 (range, 70–95)	≥5 medications	7.03%	Any cancer	Palliative radiation	30‐d mortality and OS	NS
Karuturi, 2018, U.S. 54	Population‐ based retrospective cohort	3,123 (1848 > 70 years)	Breast: 55.4% >70 years Colorectal: 63% >70 years	≥5 medications ≥11 medications	41.4% and 31.0%	Breast and colorectal cancers	Adjuvant chemotherapy	mortality	NS

^aMedian or mean, SD (range if available).

^bOdds ratio was not reported.

Abbreviations: CI, confidence interval; HR, hazard ratio; NR, not reported; NS, nonsignificant; PP, polypharmacy; OS, overall survival.

Postoperative Outcomes

See Table 1. A meta‐analysis of four studies (n = 726 patients with GI cancers) 19, 21, 22, 23 indicated a significant association between PP and postoperative complications (overall OR, 1.9; 95% CI, 1.3–2.8; p = .001; Fig. 2). A study assessing the relationship between PP and postoperative delirium found that patients with gynecologic malignancies who received five or more medications before cancer surgery were at a higher risk of postoperative delirium (OR, 1.9; 95% CI, not reported [NR]; p = .008) 24.

Figure 2.

Meta‐analysis. Forest plot for a meta‐analysis of studies evaluated the association of polypharmacy (≥5 medications) and postoperative complications (using the Clavien‐Dindo classification).

Abbreviation: CI, confidence interval.

Chemotherapy‐Related Outcomes

See Table 2. Of the 12 studies that examined the association between PP and chemotherapy‐related outcomes, 9 evaluated chemotherapy toxicity 31, 32, 33, 34, 36, 37, 39, 40, 41. The cancer types and treatments in these studies were heterogeneous. Four studies demonstrated that PP was significantly associated with severe chemotherapy toxicity 33, 36, 37, 39. In a meta‐analysis of three phase II/III trials by Woopen et al. that included 1,213 patients with advanced ovarian cancer, PP (≥5 medications) was associated with grade 3–4 hematological and nonhematological toxicities (OR, 1.1; 95% CI, NR; p < .001) 37. In a single‐center prospective study of 78 patients with breast cancer receiving first‐line chemotherapy, PP (≥5 medications) was associated with grade 3–4 toxicities (OR, 6.38; 95% CI, 2.0–23.5; p = .001) 36. In a single center retrospective study of 172 patients with solid tumors receiving irinotecan‐based therapy, the presence of any concomitant drug used to manage comorbid conditions besides cancer was associated with grade 4 neutropenia and/or grade 3–4 diarrhea (OR, 4.7; 95% CI, 1.04–21.3; p = .04) 33. PP (≥6 medications) was also associated with hospitalization (OR, 2.3; 95% CI, 1.3–3.9; p = .002) in a single center retrospective study of 318 patients with solid tumors receiving chemotherapy 40. Two additional studies did not demonstrate any association between PP and hospitalization 30, 31. PP was not associated with chemotherapy completion, dose reduction, or delay in four studies 37, 38, 40, 41.

Frailty, Falls, and Physical and Functional Outcomes

See Table 3. Seven studies examined the association between PP and frailty, falls, or physical and functional outcomes. Three studies demonstrated a positive association between PP and falls 44, 47, 48. PP was associated with impairment in either Activity of Daily Living or Instrumental Activity of Daily Living in three studies 43, 45, 46. In a single center cross‐sectional study of 385 older patients with various types of cancers (both solid and hematological), PP (≥5 medications) was associated with frailty (OR, 4.5; 95% CI, 1.9–10.5; p = NR) and prefrailty (OR, 2.4; 95% CI, 1.4–3.9; p = NR) 42.

Survival Outcomes

See Table 4. Only two studies demonstrated a positive association between PP and mortality 30, 52. In a single‐center prospective study of 83 patients with advanced ovarian cancer, PP (≥6 medications) was associated with lower overall survival (OS; OR, NR; 95% CI, NR; p = .04) 52. In another single center retrospective study of 150 patients with acute myeloid leukemia, PP (≥5 medications) was associated with increased 30‐day mortality (OR, 9.98; 955 CI, 1.18–84.13; p = NR) and overall mortality (hazard ratio [HR], 2.13; 95% CI, 1.15–3.92; p = NR) 30.

Other Outcomes

See supplemental online Table 3. Of the four studies evaluating medication adherence, three demonstrated an association of PP with adherence 39, 55, 57. In a single center prospective study of 47 patients with breast cancer, PP (≥4 medications) was associated with patients receiving nonoperative radiotherapy despite being a candidate for surgery on univariate analysis (no multivariate analysis was done; OR, NR; 95% CI, NR; p = .002) 59. In another multicenter cross‐sectional study, PP was associated with clinical depression (OR, 1.6; 95% CI, 1.1–2.3; p = .008) 60. PP was also associated with use of complementary and alternative medications (OR, NR; 95% CI, NR; p = .04) and caregiver burden (OR, 2.2; 95% CI, 1.1–4.3; p = .02) 58, 62. PP was not associated with impaired geriatric assessment domains using the Geriatric 8 screening tool, nor with radiation treatment completion or change in cancer treatment plan 49, 53, 63.

Association of PIM with Outcomes

See Table 5. Among 11 studies 25, 27, 29, 30, 31, 32, 45, 54, 58, 64, 65, three demonstrated an association between PIM and clinical outcomes. In a single center retrospective study of 171 patients with aggressive non‐Hodgkin lymphoma, PIM use by Beers criteria was associated with grade ≥ 3 toxicity (HR, 1.02; 95% CI, 1.00–1.04; p = .01), worse progression‐free survival (HR, 2.81; 95% CI, 1.4–5.8; p = .005), and higher mortality (HR, 3.12; 95% CI, 1.5–6.5; p = .003) 64. In another single center retrospective study of 475 patients (Beers criteria), PIM was associated with postoperative delirium (OR, 5.53; 95% CI, 2.02–15.10; p < .001) 27. A retrospective study that included 7,279 patients with colorectal cancer who underwent cancer surgery identified an association between PIM (using the National Board of Health and Welfare criteria) 66 and length of hospital stay (OR, 1.1; 95% CI, 1.0–1.3; p = .046) and 30‐day postoperative mortality (OR, 1.4; 95% CI, 1.1–1.9; p = .006) 29.

Table 5.

Studies examined the outcomes associated with PIM

Study	Study design	Sample size	Agea	Definition of PIM	PIM rate	Cancer type	Treatment	Outcome measures	Resultsb
Prithviraj, 2012, U.S. 45	Single center, cross‐ sectional study	117	Mean: 74.6 (range, 65–85)	Beers 2013	41%	Solid tumors	Chemotherapy	IADL and VES‐13 score	NS
Elliot, 2014, U.S. 30	Single center, retrospective cohort	150	Median: 69 (range, 61–87)	Beers 2012	19%	Acute myeloid leukemia	Induction chemotherapy	30‐d mortality, CR status, prolonged hospital stay, and ICU admission	NS
Maggiore, 2014, U.S. 31	Multicenter, retrospective cohort	500	54% ≥72	Beers, Zhan, and 2011 DAE list	Beers (29%), Zhan (11%), DAE (13%)	Solid tumors	Outpatient chemotherapy	Chemotherapy‐related toxicity and hospitalization	NS
Chiang, 2015, U.S. 65	Single center, retrospective cohort	677	Median: 72.7 (SD, 6.4)	Beers 2012	28.3%	Any	Not specified	Unplanned admission to any hospital for any reason within 30 d of discharge from the index hospital stay	NS
Nightingale, 2015, U.S. 58	Single center, cross‐sectional study	234	Mean: 79.9 (range, 61–98)	Beers 2012, STOPP, and HEDIS	51.7% (combined)	Any cancer	Any	CAM use	NS
Jeong, 2016, South Korea 27	Single center, retrospective cohort	475	Median: 76 (range, 65–96)	Beers 2012	26.7%	Any	Surgery	Postoperative delirium	OR, 5.5; 95% CI, 2.0–15.1, p < .001
Park, 2016, Korea 32	Single center, retrospective cohort	229	Median: 73 (range, 65–87)	Beers 2012	24%	Head and neck cancers	Surgery, radiotherapy, or chemo‐radiation	Grade 3–4 toxicity and hospitalization	NS
Samuelsson, 2016, Sweden 29	Population‐ based, retrospective cohort	7,279	Median: 81 (range, 75–98)	National Board of Health and Welfare criteria	22.5%	Colorectal cancer	Cancer surgery	Length of hospital stay and 30‐d postoperative mortality	Length of hospital stay: OR, 1.1; 95% CI, 1–1.3; p = .046 30‐d postoperative mortality: OR, 1.4; 95% CI, 1.1–1.9; p = .006
Choi, 2018, South Korea 25	Single center, retrospective cohort of prospective data	475	Median: 76 (range, 65–96)	Beers 2015	36%	Any	Cancer surgery	30‐d mortality and postdischarge institutionalization	NS
Karuturi, 2018, U.S. 54	Population‐ based retrospective cohort	3,123 (59% >70)	Breast: 55.4% >70 Colorectal: 63% >70	Beers 2015 and DAE list	Breast: 27.6% (Beers 2015) and 22% (DAE list) Colorectal: 24.8% (Beers 2015) and 15.5% (DAE list)	Breast and colorectal cancers	Adjuvant chemotherapy	Emergency room visit, hospitalization, and death	NS
Lin, 2018, U.S. 64	Single center, retrospective cohort	171	Median: 70 (range, 65–77)	Beers 2015	47%	Aggressive non‐Hodgkin lymphoma	Chemo‐immunotherapy	Treatment delay, dose reductions, grade 3–4 toxicity, PFS, OS	Grade 3–4 toxicity: HR, 1.02; 95% CI, 1–1.04; p = .014 PFS: HR, 2.8; 95% CI, 1.4–5.8; p = .005 OS: HR, 3.1; 95% CI, 1.5–6.5; p = .003 Other outcomes: NS

^aMedian or mean, SD (range if available).

^bMultivariate unless otherwise specified.

Abbreviations: CAM, complementary alternative medicine; CI, confidence interval; CR, complete response; DAE, drugs to avoid in the elderly; HR, hazard ratio; IADL, instrumental activity of daily living; ICU, intensive care unit; NS, nonsignificant; OR, odds ratio; OS: overall survival; PFS, progression‐free survival; PIM, potentially inappropriate medications; VES, Vulnerable Elders‐13.

Discussion

To the authors’ knowledge, this is the first systematic review that summarizes the associations of PP/PIM with outcomes in older adults with cancer. Included studies were heterogeneous in terms of study design, study population, sample size, PP definitions, and outcomes examined. The wide range of definitions used contributes to the wide range of PP prevalence reported, from 2% 56 to 80% 45. PIMs were assessed using the Beers criteria in the vast majority of the studies, which is the most commonly used tool for evaluation of PIMs in both clinical and research settings; other tools supplemented the use of the Beers criteria in some studies.

The meta‐analysis shows that PP is associated with postoperative complications using the Clavien‐Dindo classification. In addition, several studies suggest that PP is associated with chemotherapy toxicity, frailty, falls, and medication nonadherence. Most studies did not show an association between PP and survival, and no studies showed an association between PP and chemotherapy completion. PIM is associated with postoperative complications (delirium and readmission), and two studies indicate that PIM may be associated with higher mortality and lower progression‐free survival 29, 64.

Most studies did not show an association between PP and survival, and no studies showed an association between PP and chemotherapy completion. PIM is associated with postoperative complications (delirium and readmission), and two studies indicate that PIM may be associated with higher mortality and lower progression‐free survival.

Other reviews in the general geriatric population have found a positive association of PP with functional decline 67, 68. However, this is the first systematic review to summarize the association between PP and postoperative outcomes and chemotherapy toxicity in older adults with cancer. Because of increased frailty and geriatric syndrome burden in older adults with cancer, as well as the contribution of chemotherapy and supportive care regimens to the overall number of medications, data from general older adult populations are unlikely to be generalizable.

Four studies assessing the impact of PP on postoperative outcomes used a similar population (patients with gastrointestinal cancers), the same definition of PP (≥5 medications), and the same outcome (postoperative complications using the Clavien‐Dindo classification). A pooled analysis of these studies demonstrated that PP was associated with postoperative complications, with an overall OR of 1.9 (95% CI, 1.3–2.8). This may be due to an increased risk of adverse drug events and drug‐drug interactions in the presence of anesthetic and other perioperative medications such as analgesics and antibiotics. In addition, hospitalization itself is associated with an increased risk of postdischarge medication‐related adverse events 69. We are unable to determine which medications or class of medications were most associated with postoperative complications; medications assessed as being potentially inappropriate by Beers criteria or other PIM measures have shown associations with postoperative length of stay, mortality, and postoperative delirium across several studies 27, 29, but the effect of PIM on postoperative complications in older patients with cancer is unknown.

Similarly, the association between PP and chemotherapy outcomes remains unclear. Several studies demonstrate that PP is associated with grade ≥ 3 chemotherapy toxicity, but other studies failed to show an association with chemotherapy dose intensity or early discontinuation of therapy. It remains unclear whether PP affects receipt of chemotherapy and, in turn, cancer‐related survival. This question may be most critical in older patients being treated with curative intent, for whom chemotherapy dose delays or reductions may substantially affect survival outcomes. PP increases the risk of clinically relevant drug‐drug interactions which may potentiate chemotherapy toxicity and/or adverse drug events 70.

PP was found to be significantly associated with reduced OS in only 2 out of 11 studies 30, 52. The lack of association in other studies may be due to the advanced stage of cancer and poor overall prognosis of the included patients. In very sick patients, PP may be appropriate and may serve to prolong survival 71. Competing risk of cancer mortality is also possible: patients with advanced cancer may die from their cancer before the adverse effects of PP accrue.

Among the other outcomes evaluated, PP was associated with improved adherence to adjuvant endocrine therapy 55, except among patients frequently using opioid‐containing analgesics, anxiolytics or antipsychotics, and antidepressants who had lower adherence to their cancer therapy. This suggests that certain drug classes may have a disproportionate effect on outcomes and that a simple count of medications to assess PP may incorrectly assess risk. Adherence is an important predictor of clinical outcomes 72, and more data are needed to assess the effects of PP, PIM, and medication burden on adherence, particularly for the increasing number of oral chemotherapy agents 73. It is also possible that medication burden could impair patients’ ability to adhere to supportive care regimens, thereby adversely affecting outcomes, but no data specifically evaluates this hypothesis.

There are limited studies evaluating the association of PIM with clinical outcomes, and most did not show significant associations. Existing criteria and tools to determine PIM are primarily derived from the general geriatric population 74, 75, and the applicability of these tools in the setting of cancer is unclear. The time frame for identification of PIM in patients with cancer may be problematic: many are prescribed supportive care medications considered potentially inappropriate in older adults, although these are usually administered transiently and may be specifically appropriate to treat symptoms related to cancer or cancer treatment. One study has noted a transient elevation in PIM prevalence after a lung or colon cancer diagnosis, which was mostly due to the use of supportive care medications 76. Most current PIM assessment tools consist of explicit criteria which account for little patient context; assessment of the appropriateness of medications could be improved with tools using implicit criteria, such as the MAI, but such tools require time and expertise for application. The MAI assesses the appropriateness of all medications on a patient's list: any medication may be inappropriate within a certain context (such as lack of an ongoing indication for the medication or lack of time to benefit based on life expectancy).

Two population‐based studies were published after our search and were not included in our systematic review. The first found an association of PIM (using Beers criteria) with greater health care utilization and higher health care costs in a cohort of 17,630 older patients with breast, colorectal, and prostate cancers 77. Another study of 3,123 older patients with breast and colorectal cancers did not find an association between PIM (using STOPP criteria) and emergency department visits, hospitalization, or death 78.

Despite some limited data suggesting that PP/PIM can affect outcomes in older adults with cancer, it is unknown whether intervening on PP or PIM can improve outcomes. There is growing interest in “deprescribing” interventions, involving planned withdrawal of medications. Deprescribing has been shown in preliminary studies to reduce the number of PIMs, falls, and mortality in certain populations. Although deprescribing has been shown to be feasible in older adults with cancer, data are lacking about its efficacy 79, 80.

Deprescribing has been shown in preliminary studies to reduce the number of PIMs, falls, and mortality in certain populations. Although deprescribing has been shown to be feasible in older adults with cancer, data are lacking about its efficacy.

More studies are needed to determine the prospective outcomes of targeted deprescribing interventions (whether decreasing PP will improve outcomes of interest) as well as determine how these interventions can be implemented for older adults with cancer, including who should deliver them (oncologist, pharmacist, or other health care provider).

This review underscores the limited data available to assess the impact of PP and PIM on outcomes for older adults with cancer. Prospective data are limited; most studies are retrospective cohort or cross‐sectional studies. Many studies were not designed with the specific objective to evaluate the effects of PP and PIM on outcomes. Many studies included PP and/or PIM as one of many covariates assessed (in the setting of geriatric assessment, for example), rather than as the primary variable of interest; this approach may be suboptimal for confounding controls and may lead to misinterpretation and lack of reproducibility 81. The studies are not consistent in their definitions of PP, and most of the included studies did not specify how they counted the number of medications. Many medications contributing to the risk of adverse outcomes are over‐the‐counter (e.g., diphenhydramine) or are typically prescribed on an as needed basis (e.g., benzodiazepines). Omission of these groups of medications from the assessment could limit the validity and applicability of the data. In addition, patients who are on more medications are more likely to have more comorbidities and/or functional decline, which are important confounders to consider as they may lead to adverse outcomes; although some papers adjusted for these factors, it was not always possible to determine the independent contribution of PP and PIM. Finally, included studies were not limited to those enrolling only older adults, most did not assess the potential for interaction between age and PP and PIM, and studies included subjects with variable cancer types and other characteristics, which may carry different risks of adverse outcomes.

Our study has several additional limitations. We did not include non‐English publications, and six studies were not identified in the initial search, so it is possible that other studies were missed. Most of the studies were perceived to have poor to fair quality in relation to the specific objective of this review. The majority were retrospective or cross‐sectional in design, and the causal relationships between PP and the various outcomes cannot be determined. Most studies (27/47) evaluated PP in the context of a geriatric assessment (GA) and included limited information on the medications (e.g., classes, doses). We were unable to determine if the results of the GA were available to treating physicians; GA is known to influence discussion about medications and may drive medication changes, which in turn may affect outcomes of interest 82. Although the majority of the included studies provided quantitative definition of PP, only about one‐third clearly specified whether prescription, supplemental, and over‐the‐counter medications were included.

Conclusion

This is the most comprehensive review to date assessing associations between PP and/or PIM and health outcomes in older patients with cancer. PP and PIM are prevalent in older adults with cancer, but definitions are very heterogeneous, complicating interpretability of associations with outcomes of interest. PP is associated with postoperative complications, functional impairment and possibly chemotherapy‐related toxicity, although prospective studies with detailed medication reviews are needed to further investigate these associations. Data are very limited for associations with PIM and outcomes in older patients with cancer, and widely‐used PIM measures may not be as useful in this population. Clear and validated definitions and instruments are needed to investigate PP and PIM in older adults with cancer and to develop interventions, such as deprescribing interventions, to improve outcomes for these vulnerable patients.

Author Contributions

Conception/design: Mostafa R. Mohamed, Erika Ramsdale, Kah Poh Loh, Supriya G. Mohile

Collection and/or assembly of data: Mostafa R. Mohamed, Kah Poh Loh, Asad Arastu, Spencer Obrecht

Data analysis and interpretation: Mostafa R. Mohamed, Erika Ramsdale, Kah Poh Loh, Huiwen Xu, Daniel Castillo

Manuscript writing: Mostafa R. Mohamed, Erika Ramsdale, Kah Poh Loh, Manvi Sharma

Final approval of manuscript: Mostafa R. Mohamed, Erika Ramsdale, Kah Poh Loh, Asad Arastu, Huiwen Xu, Spencer Obrecht, Daniel Castillo, Manvi Sharma, Holly M. Holmes, Ginah Nightingale, Katherine M. Juba, Supriya G. Mohile

Disclosures

The authors indicated no financial relationships.

Supporting information

See http://www.TheOncologist.com for supplemental material available online.

Supplemental Appendices

Supplemental Tables