Why Medicines Pile Up — An Introduction to Polypharmacy and Time to Benefit

Five Guidelines, About Twelve Drugs

A 79-year-old woman lives with chronic obstructive pulmonary disease, type 2 diabetes, osteoporosis, hypertension, and osteoarthritis. She is not a real person, but rather a hypothetical patient constructed by researchers in a review published in 2005. The researchers looked up the clinical practice guidelines for these five conditions, which are common in primary care in the United States, and added together what each guideline recommended for this single individual. Following all the guidelines would saddle this person with about 12 medications (costing 406 dollars a month in drugs) along with complex non-drug therapies, and created the potential for harmful interactions between drugs and between drugs and diseases (PMID 16091574). In the same review, most guidelines failed to address how to apply recommendations to older adults with multiple conditions, and usually omitted the burden on patients and caregivers, short- and long-term goals, and the quality of evidence (PMID 16091574). The authors concluded that caring for older adults with multiple conditions according to the guidelines of the time could lead to undesirable outcomes (PMID 16091574).

Multimorbidity is a condition in which a person has two or more long-lasting diseases at the same time. A perspective paper from New Zealand notes that there are now more people with multimorbidity than people with only a single long-lasting disease, yet guidelines still treat diseases one by one (PMID 29565938). According to the same paper, combining recommendations for each disease in one person frequently causes interactions between treatments and a heavy treatment burden, people with multimorbidity are often excluded from the trials that form the basis of guidelines, and the treatment goals of guidelines are frequently out of step with the patient's goals (PMID 29565938). Looking solely at a long prescription list does not tell us who is at fault. Each line might have been justified when viewed by itself. Even if each individual guideline is correct, a prescription list that adds them all up for one person is an object that no single guideline has evaluated.

The earlier articles in this series read a single test, a single indicator, and a single threshold in front of an individual patient. This article looks at what happens when those numbers pile up layer by layer on one person. There are two axes. One is the path by which the prescription list grows, and the other is the time attached to each line of the list—in other words, the time it takes for the benefit of a preventive drug to arrive. How survival analysis handles the thought that a different ending may arrive first within that time was covered in "When People Don't Stay in Line — The Ones Who Shifted, and the Ones Who Left for Another Reason". Here, we use that idea as a framework for prescribing decisions.

The Line of Five

In older samples, the proportion of people taking multiple medications together ranges from 2.6% to 86.6% depending on the study (PMID 38452787). Polypharmacy is a state where a person uses multiple drugs together, commonly taking five or more as the cutoff. There is no academic consensus on this threshold, and five is simply the most frequently used number (PMID 31732240). Depending on the period over which drugs are counted, the definition varies between medications used together at a given moment, medications accumulated over a period, and medications taken continuously, and in studies gathered by one review, the prevalence varied from 4% to 57% depending on the definition (PMID 31732240). Looking across 87 gathered studies, most used the same line of two or more diseases and five or more drugs, but what was counted as a disease and a drug differed widely (PMID 38452787). As we saw in "A Line Painted on a Slope — An Introduction to Reference Ranges and Diagnostic Thresholds", five is also a line drawn on a continuous axis of the number of drugs.

Polypharmacy is associated with an increased risk of adverse drug reactions, cognitive and functional decline, frailty, falls, hospitalization, and death (PMID 38719530). One review highlights the number of prescribed drugs as the single best factor predicting the risk of inappropriate prescriptions and adverse drug reactions in older adults (PMID 25798731). However, authors of another review believe that confounding by multimorbidity—meaning the portion where the diseases themselves that made people use many drugs lead to poor outcomes—has been underestimated so far (PMID 30540223). From the mere observation that people with many drugs have many poor outcomes, we cannot say that removing drugs improves outcomes. To answer that question, trials that try reducing them are needed, and those trials are examined later.

The criticism that counting alone is not enough is repeated across reviews. Examining whether those drugs are appropriate for an individual tells us more than counting the number of pills (PMID 39475037). The same review lists instances where prescriptions become inappropriate: when there is no evidence, when the likelihood of harm outweighs benefit, when dangerous drugs or combinations are included, when the patient experiences treatment burden or reduced adherence, and when another drug is added for a side effect (PMID 39475037). It also notes that guidelines for individual diseases recommend prescribing but usually leave out stopping criteria, and barriers that hinder reducing medications cause drugs to pile up over time (PMID 39475037). Here is one answer to the title's question of why drugs accumulate. Rules for adding exist for every disease, while rules for subtracting are rare.

Potentially inappropriate medications are drugs where the risks in older adults may outweigh the benefits, so they are generally better avoided. The American Geriatrics Society Beers Criteria is a list specifying such drugs, noting medications that older adults should generally avoid in most situations or with specific diseases and conditions (PMID 37139824). This criteria was created for use in adults aged 65 and older across outpatient, inpatient, and residential settings, but excludes hospice and end-of-life care, was designed for the United States, and states itself that it should be used to support rather than replace shared decision-making between clinicians and patients (PMID 37139824). Being on the list is a sign to 「examine closely」, and does not give the same answer for everyone taking that drug.

The burden of medication does not land on the body alone. The treatment burden borne by a person with multimorbidity consists of tasks such as medical visits and taking medications, along with the social, emotional, and economic impacts those tasks produce; when this burden grows, people fail to adhere to treatment, diseases progress, quality of life drops, and the burden on caring family members also increases (PMID 39342121). The length of the prescription list is itself an outcome.

When a Side Effect Becomes a New Diagnosis

One detail was almost missing from the medical records of 120 US veterans aged 66 and older who were newly prescribed a loop diuretic after starting a gabapentinoid (gabapentin or pregabalin) (PMID 41329486). Among them, 88.3% were already taking five or more long-term medications. Almost all of the physicians recorded the reason for using the diuretic, and the most common reason was leg swelling, yet records listing a gabapentinoid as a candidate cause for the swelling were found in only 4 patients (3.3%) (PMID 41329486). The most frequently listed candidate was heart failure (39.2%).

A prescribing cascade is the process of misinterpreting a drug's side effect as a new disease and adding another medication to treat it. Prescribing cascades are recognized as one of the pathways that increase polypharmacy (PMID 40171689). Well-known examples include drugs with anticholinergic effects causing cognitive decline, indigestion, and constipation, which then lead to prescriptions for dementia medications, acid suppressants, and laxatives, while swelling caused by calcium channel blockers and gabapentinoids leads to unnecessary diuretics (PMID 40171689). The prevention strategies cited in this review are carefully re-examining side effects, reducing medications when appropriate, and training healthcare staff on how to interpret symptoms (PMID 40171689).

At the end of the cascade comes the harm of the new diuretic. In the aforementioned veteran records, within 60 days of starting the diuretic, 28 patients (23.3%) experienced 37 events that could have been due to the diuretic, with the common ones being worsening kidney function in 9 patients (7.5%), orthostatic hypotension in 7 patients (5.8%), electrolyte abnormalities in 6 patients (5.0%), and falls in 5 patients (4.2%), and 6 patients (5.0%) received emergency room or hospital care (PMID 41329486). The authors concluded that downstream harms potentially resulting from this overlooked cascade are common (PMID 41329486). However, because this was a chart review without a comparison group, it cannot be confirmed that these events were caused by the diuretic, and the authors also counted them as events that 「could have been」. The same pattern appears in administrative data. Among 28,952 elderly dementia patients in Nova Scotia, Canada, starting a medication with anticholinergic burden was associated with a high risk of starting an acid suppressant within 180 days (hazard ratio 1.38), and cases defined as a cascade were 1,845 patients (6.4%) (PMID 35814221). Because this association was observed through prescription data, this data does not tell us whether individuals actually experienced side effects.

Not all cascades are wrong. Twelve geriatric pharmacotherapy experts used a Delphi consensus to filter 107 candidates and created a list of 65 「potentially inappropriate prescribing cascades」, and the starting point of this work was distinguishing between inappropriate cascades and those that might be appropriate (PMID 40742670). Choosing to add a medication to manage a side effect knowingly because the benefit of the original drug is substantial can be a valid option. What becomes a problem is when that choice is made without knowing. The first question when facing a new symptom is 「Could this symptom be due to a medication I am currently taking?」, and skipping that question adds another line to the list.

When the Benefit Arrives

When the survival curves from 8 randomized trials of statins in people without cardiovascular disease involving 65,383 participants were re-combined, it took 2.5 years (95% confidence interval 1.7 to 3.4 years) to treat 100 people and prevent the first major cardiovascular event (PMID 33196766). In the same analysis, it took 1.3 years to prevent 1 event per 200 people, and 0.8 years per 500 people (PMID 33196766). The answer to "When does the benefit arrive?" depends on "How much benefit are we waiting for?"

Time to benefit is the time it takes for a statistically significant benefit to first appear between the drug group and the untreated group in a trial. A review outlining this concept also defines the time to harm—the time until harm appears with statistical significance in the same manner—and notes that measuring both times requires clearly defining what counts as a benefit and a harm as patient-important outcomes (PMID 23749475). The same review points out that methods to extract this information from current trial data are not established, and that clinicians' lack of this information is an urgent challenge for trial design and reporting (PMID 23749475). The aforementioned statin meta-analysis measured the point at which the absolute risk reduction reached a pre-specified magnitude instead of statistical significance. This is another reason why the numbers vary depending on the definition.

The average age in those statin trials was 55 to 69 years, the average follow-up was 2 to 6 years, and only 1 out of the 8 trials reduced all-cause mortality (PMID 33196766). The authors concluded that for ages 50 to 75 with a life expectancy of at least 2.5 years, statins can help prevent a first major cardiovascular event, but there is no evidence that they reduce mortality (PMID 33196766). Both halves of this conclusion must remain intact. In a meta-analysis of 10 randomized trials involving 85,829 participants restricted to people without cardiovascular disease, lipid-lowering drugs did not reduce all-cause mortality (3.9% vs. 3.9%), but they did reduce major cardiovascular events (relative risk 0.76), myocardial infarction (0.64), stroke and transient ischemic attack (0.75), and cardiovascular hospitalization (0.75) (PMID 41864911). The benefit of primary prevention statins lies mainly in avoiding myocardial infarction and stroke while alive, and that benefit takes time to arrive.

We cannot assume that benefits arrive later in older age. A post-hoc re-analysis of a trial comparing the osteoporosis drug alendronate and a placebo in postmenopausal women calculated the time to benefit for clinical fractures, finding it was 11 months overall, 8 months in those aged 70 and older, and 19 months in those under 70 (PMID 26884390). The absolute risk reduction was also larger in those 70 and older at 1.4%, compared to 0.7% in those under 70 (PMID 26884390). Because this is a post-hoc re-analysis of a single trial, the numbers themselves cannot be generalized, but in this trial, the benefit arrived sooner in the older group. Based on this trial population, the authors concluded that prescribing alendronate to those 70 and older is useful, and recommended that trials report the time to benefit (PMID 26884390).

Time to benefit is the average of the trial population and does not measure the time remaining for an individual. The figures of 2.5 years or 8 months come from trials following people of a specific age group and risk profile over a specific period. Attaching this time to each line on a prescription list provides one more piece of material to use in judgment. The judgment itself incorporates both the time remaining for that person and their goals.

A Different Ending That Comes First

There is almost no direct randomized evidence on the problem of using statins in people over 80 without atherosclerotic cardiovascular disease (PMID 42706978). Systematic reviews addressing this problem cite reasons why results from younger groups are difficult to transfer: competing mortality, frailty, multimorbidity, polypharmacy, and the short remaining time for benefits to accumulate, all of which grow larger as age increases (PMID 42706978).

A competing risk is a risk where an event other than the one of interest happens first, eliminating the very chance for the event of interest to occur. If a person passes away from another illness before the myocardial infarction that statins try to prevent can arrive, the benefit of the statins never reaches that person. The burden and harm of taking the medicine remain just the same during that time. The time to reach benefit is a value determined by the drug and the risk profile of the trial group, and the likelihood that another ending will arrive first within that time is a trait possessed by that individual. A prescribing decision is the act of bringing these two face-to-face.

In that systematic review covering people 80 and older, the evidence was scattered like this: trials extending up to age 82 mixed primary and secondary prevention, one trial failed to show benefit in those 75 and older, and two trials that reduced cardiovascular events in those 70 and older included few people in their 80s (PMID 42706978). In observational cohorts, statin users had fewer deaths and cardiovascular events, but those estimates were limited by the bias that healthy people tend to take their pills more reliably and by residual confounding (PMID 42706978). The authors summarized that evidence supports benefit up around age 75, but beyond age 80 the evidence is indirect and of low or very low certainty (PMID 42706978). The authors' judgment is that the people for whom benefit is most plausible are healthy octogenarians with high cardiovascular risk and sufficient life expectancy, and that routinely starting it anew for frail people with limited prognosis is unlikely to provide meaningful benefit (PMID 42706978). The conclusion was that there is insufficient evidence to routinely start it for all people 80 and older, and that while waiting for the results of ongoing trials, one should determine it together by weighing baseline risk, frailty, life expectancy, time to reach benefit, treatment burden, drug interactions, and the patient's goals (PMID 42706978).

Competing risks also enter models that weigh benefits and harms. A modeling study that weighed the benefits and harms of statins over 10 years targeting ages 40 to 75 incorporated the frequency of harms, people's preferences regarding outcomes, and the competing risk of dying first from non-cardiovascular causes, and calculated the 10-year cardiovascular risk threshold at which the probability of a net benefit becomes 60% or greater (PMID 30508425). That threshold was 14% at ages 40–44 and 21% at ages 70–75 in men, and 17% and 22% in women (PMID 30508425). It is a modeling study, and it has the limitation that some harms lacked age-specific data (the results were similar under most other assumptions as well). The authors concluded that net benefit occurs at higher risks than those set by most guidelines, and that the threshold varies greatly by age, sex, and type of statin (PMID 30508425). In this model, even with the same drug, older age required a higher risk for a net benefit, and that calculation incorporated baseline risk, harms, preferences, and competing risks together.

Frailty is not necessarily the extinction of benefit. In a secondary analysis of a blood pressure trial involving 4,733 patients with type 2 diabetes (average 62.7 years old, 39.9% frail), within the standard glycemic control group, aggressively lowering systolic blood pressure reduced a composite of myocardial infarction, stroke, and cardiovascular death (hazard ratio 0.75), and that benefit was consistent regardless of frailty (PMID 39915072). In the intensive glycemic control group, the benefit of intensive blood pressure control was not seen (hazard ratio 1.08) (PMID 39915072). The authors concluded that for type 2 diabetes patients receiving standard glycemic control, frailty should not become a reason to withhold intensive blood pressure control (PMID 39915072). The framework for considering competing risks does not tilt only toward removing medications. The mere fact of being frail may not serve as a basis to withdraw a treatment whose benefit has been confirmed, as it was in this trial.

Trials That Tried Less

381 adults with an estimated remaining life expectancy of one month to one year and recent functional decline were randomly assigned to either stop or continue statins (PMID 25798575). The researchers who conducted this trial started from the recognition that for patients with limited prognosis, especially when benefits take years to accumulate, the risks of some drugs may outweigh the benefits (PMID 25798575).

Deprescribing is the planned process of reducing or stopping drugs where harms may outweigh benefits, and monitoring the outcome. One review suggests this process in five steps (PMID 25798731).

1. Review all the medications a patient is currently taking and identify the reason for each.
2. Weigh the patient's overall risk of harm from medications to determine how aggressively to deprescribe.
3. Compare the current and future benefits of each drug against its current and future harms and burdens.
4. Prioritize drugs starting with those where harms most outweigh benefits, and where stopping carries the lowest risk of withdrawal or disease rebound.
5. Execute the tapering plan and closely monitor for improvements or worsening.

The "future benefits" in the third step include the time to benefit and competing risks. The fifth step means that deprescribing is also an intervention whose results must be monitored.

Participants in the statin discontinuation trial had a mean age of 74.1 years, with 22.0% having cognitive impairment and 48.8% having cancer (PMID 25798575). The proportion who died within 60 days was 23.8% in the stopped group and 20.3% in the continued group, showing no significant difference (90% confidence interval -3.5 to 10.5 percentage points), but the trial's predefined non-inferiority margin was not met (PMID 25798575). The stopped group had slightly higher total quality-of-life scores (7.11 vs 6.85), used fewer non-statin medications, and had few cardiovascular events in both groups (13 vs 11 cases) (PMID 25798575). The authors concluded that in this situation, stopping statins is safe and can be associated with benefits such as improved quality of life, and recommended discussing the uncertain benefits and potential quality-of-life decline of continuing with the patient (PMID 25798575). Both the authors' conclusion and the unmet non-inferiority margin are results of this trial.

In a systematic review combining 36 studies comparing statin discontinuation versus continuation, this was the only randomized trial. Based on this trial, that review interpreted that stopping likely makes no difference to 60-day mortality in people with a life expectancy of under a year (PMID 39051828). In the same review, non-randomized studies looking at other populations consistently suggested that stopping could be associated with increased mortality (hazard ratio 1.92, 9 studies), cardiovascular death (1.63), and cardiovascular events (1.31), with similar results even in those aged 75 and older, though methodological limitations left high uncertainty (PMID 39051828). Trial results from people near the end of life cannot be transferred to people with longer life expectancies. Nor are the signals from non-randomized studies definitive. The two sets of results are evidence of different strengths for different populations.

Patient-side concerns were smaller than expected. When 297 cognitively intact participants from the same discontinuation trial were surveyed, less than 5% worried that discontinuation meant their doctor was giving up on them, 18% each reported being told to take the drug for life and feeling that their past efforts were wasted, and many cited lower medication costs (63%) as an advantage (PMID 28520522).

For blood pressure medications, the reduction strategy itself was tested. Across 69 UK primary care practices, 569 people aged 80 or older with a systolic blood pressure under 150 mmHg, taking two or more blood pressure medications, and deemed suitable for dose reduction by their primary care physician, were divided into a group where one blood pressure medication was withdrawn and a group receiving usual care (PMID 32453368). After 12 weeks, the proportion with a systolic blood pressure under 150 was 86.4% versus 87.7%, meeting the predefined non-inferiority margin (PMID 32453368). However, the systolic blood pressure in the withdrawal group was higher by a mean of 3.4 mmHg, 66.3% of people maintained the reduction even after 12 weeks, and serious adverse events showed no statistical difference at 4.3% versus 2.4% but were numerically higher in the withdrawal group (relative risk 1.72, 95% confidence interval 0.7–4.3) (PMID 32453368). The authors wrote that further research is needed to understand long-term clinical outcomes (PMID 32453368). What this trial showed goes only as far as demonstrating that blood pressure control did not drastically collapse over 12 weeks in selected octogenarians.

When We Pool Them — Fewer Pills, Uncertain Outcomes

Even when pooled together, reducing medications in older adults did not significantly lower mortality. In a 2024 meta-analysis pooling 259 studies of reducing one or more medications in older adults, the odds ratio for mortality was 0.96 (95% confidence interval 0.84 to 1.09) in randomized studies and 0.70 (95% confidence interval 0.36 to 1.38) in non-randomized studies, neither of which was significant (PMID 39164070). In the same analysis, mortality was significantly reduced in the randomized subgroups aged 65 to 79 (odds ratio 0.71) and in interventions tailored to the individual patient (0.79) (PMID 39164070). The authors concluded that prescriber-led medication reduction can be achieved alongside the potentially important benefit of improved survival, particularly when patient-tailored interventions are initiated in early old age (PMID 39164070). Because the overall estimate was not significant, this conclusion relies on subgroups, and subgroup results provide weaker evidence than the overall result. In the same research team's 2016 edition, non-randomized studies appeared to substantially reduce mortality (odds ratio 0.32) while randomized studies did not (0.82) (PMID 27077231), whereas in the 2024 edition with more accumulated studies, the non-randomized side was no longer significant either.

The picture becomes clearer when combining 118 randomized trials involving 417,412 participants that targeted inappropriate prescriptions in older adults in primary care (PMID 40577011). The intervention reduced prescribed medications by about 0.5 per person, but showed no clear difference in mild adverse drug reactions (relative risk 0.92), injurious falls, quality of life, emergency department visits, hospitalizations (0.95), or all-cause mortality (0.94, 95% confidence interval 0.85 to 1.04) (PMID 40577011). The authors noted that such interventions can safely reduce the number of medications in older adults, while recommending that core outcomes such as quality of life, hospitalization, and mortality continue to be reported consistently in the future (PMID 40577011).

Narrowing the scope reveals other signals. In a meta-analysis focusing on community-dwelling older adults, comprehensive medication reviews may have reduced all-cause mortality (odds ratio 0.74), but the certainty of the evidence was low and the effect on falls, quality of life, and hospitalization was assessed as little to none (PMID 32820421). Conversely, combining 15 studies that discontinued preventive medications—such as blood pressure lowering drugs, statins, anticoagulants, and diabetes medications—in older adults with advanced frailty, dementia, or limited life expectancy yielded a relative risk for mortality of 1.15 (95% confidence interval 0.98 to 1.35), which was not a significant increase, though there was substantial heterogeneity across studies (I² 93%) and very low certainty (PMID 41840386). In this analysis, discontinuation was not associated with hospitalization, falls, fractures, or worsening quality of life, and major adverse cardiovascular events were not significantly increased, though the range was wide (relative risk 1.37, 95% confidence interval 0.70 to 2.70), while discontinuing blood pressure medications slightly increased systolic blood pressure (PMID 41840386). In a randomized trial targeting central nervous system-active drugs to reduce falls, the rate of discontinuing tricyclic antidepressants increased, but falls requiring treatment over 18 months did not decrease (hazard ratio 1.11) (PMID 39052289).

While deprescribing is consistent in reducing pill counts, the evidence that it reduces mortality or hospitalization remains uncertain. The result that pill counts decrease also requires an assessment of certainty. A Cochrane review evaluated whether pharmaceutical care interventions reduce the number of potentially inappropriate medications as uncertain, and rated the certainty of that evidence as very low (PMID 37818791). Signals of an association with adverse outcomes were also not clear, but the certainty of that judgment varies across populations. It was very low for frail individuals or those nearing the end of life (PMID 41840386), whereas an analysis pooling 118 primary care randomized trials reported no clear difference in hospitalization or mortality without grading certainty (PMID 40577011). One review noted that deprescribing studies fail to demonstrate benefits in hard clinical outcomes due to differences in intervention intensity, poor execution, lack of collaboration between pharmacists and prescribers, and approaches that fail to reflect patient priorities and multimorbidity, while also writing that high-intensity multidisciplinary interventions involving clinical pharmacologists hold promise for reducing hospitalization and adverse drug reactions (PMID 41099738). The question of deprescribing does not close with a simple "Is it safe?", but breaks down into whose medications were reduced, which drugs, and in what manner.

A Person's Prescription List

Let us apply the thoughts so far to a hypothetical person one by one. She is a 78-year-old woman with hypertension, type 2 diabetes, osteoporosis, and knee osteoarthritis. This is the composition of the 79-year-old woman in the first section minus chronic obstructive pulmonary disease. To avoid reading like a prescription, drugs are listed only by class, and dosages are not written.

Opening the guidelines for each disease adds lines to the prescription list. Drugs go up for blood pressure, blood sugar, bones, and knee pain, respectively, and a drug to prevent a first cardiovascular event also enters as a candidate. I will not fabricate the exact number of lines in this composition. The calculation in the first section adding five diseases was 12 items (PMID 16091574). When a list reaches five lines, this person falls under polypharmacy by the definition most studies use (PMID 31732240). That fact is a marker associated with the risk of bad outcomes occurring in this person, but how much of that association is due to drugs must be parsed separately (PMID 30540223).

Months later, her ankles swell. If this person's blood pressure drug belongs to the calcium channel blocker class and the edema is due to that drug, adding a diuretic to manage this edema becomes a named prescription cascade (PMID 40171689). Edema also has other causes, and the candidate most frequently recorded in veterans' records was heart failure (PMID 41329486). What was missing from those records was the candidate called a drug. The primary question at this spot, before what to manage the edema with, is whether this edema could be due to a drug she is taking right now.

Now, let's attach arrival times to each line of preventive medicine. Among osteoporosis drugs, alendronate showed a fracture benefit in women aged 70 and older at 8 months in a post-hoc analysis of a trial (PMID 26884390). Statins, which prevent first cardiovascular events, took 2.5 years per 100 people in trials of people aged 50 to 75 (PMID 33196766), but the two times are measured differently (the point where the difference in the number of fractures exceeded the statistical control limit versus the point where the absolute risk reduction reached 1%) so they cannot be compared with each other. And this person is 78, which is outside the age range of the statin trials. Evidence supports benefit up to around age 75, and past age 80 the evidence is indirect (PMID 42706978), so this person's age is in between. In an analysis of trials of people with the same diabetes, the benefit of intensive blood pressure control was consistent regardless of frailty when receiving standard blood sugar control (PMID 39915072). However, because it was a trial with an average age of 62.7, it is a much younger group than this person. The time attached to each line and the strength of the evidence are all different.

What creates the crossroads are the time remaining and the goals for this person. If this person walks, manages daily life, and has no other life-threatening diseases, the room for competing risks to block the benefits of preventive medicine is relatively small. A review addressing the issue of starting a statin anew in octogenarians also named healthy people with high cardiovascular risk and sufficient life expectancy as those for whom the benefit is most plausible, but the evidence at that age is indirect and has low certainty (PMID 42706978). In this branch, competing risks are not a reason to remove the statin line. However, whether this person's cardiovascular risk is high must be examined separately, and drugs that need to be re-examined for other reasons, such as a prescription cascade or potentially inappropriate medication, still remain. Non-randomized studies looking at groups outside of those nearing the end of life suggested an association between stopping statins and bad outcomes, though uncertain (PMID 39051828). Conversely, if this person develops an advanced disease so that the remaining life expectancy is estimated to be under a year and function is declining, this person approaches the participation criteria of a statin discontinuation trial (PMID 25798575). In that trial, 60-day mortality in the discontinued group was not significantly different and quality of life was slightly better, but non-inferiority was not proven (PMID 25798575). The results of pooling studies that stopped preventive medications in such people also go only as far as "not associated with bad outcomes, certainty is very low" (PMID 41840386).

The final question in both branches is what this person wants to avoid the most. When 207 people aged 60 and older with hypertension and two or more other chronic diseases were asked, the outcome they worried about most was stroke, and the outcome they worried about least was treatment burden, but for no outcome did more than 25% of respondents consistently choose it as what they worry about most or least (PMID 31856842). The authors concluded that no outcome should be ignored in guidelines and economic evaluations (PMID 31856842). Removing a line takes on the risk of the event that line was blocking, and keeping a line takes on the burden and harm. Shared decision-making is a process where medical staff and patients put the best evidence together with the patient's goals and prognosis, weigh the pros and cons of each treatment, and decide on a treatment plan together. Both the octogenarian statin review and the Beers criteria pointed to this process at the end (PMID 42706978; PMID 37139824).

This scene does not draw a conclusion. What this framework asks is not "Shall we reduce medications?", which comes later, but "Does this line have time to deliver its benefit to this person?" That answer differs for each person and each line, and for many lines the answer is "Yes".

Where Else to Look

Trial information showing time-to-benefit and time-to-harm is still scarce. Recommendations exist to report these times in every trial (PMID 26884390), but the methods to extract them from current trial data were not established at the time of a 2013 review (PMID 23749475). Whether deprescription changes hard outcomes like death, hospitalization, or quality of life, and if so, for whom and through what method, also remains an open question (PMID 41099738).

Articles to read next along with this series are as follows. How to read numbers like 1valence per 100 people as absolute risk is in What Does "50% Lower Risk" Mean in People? — Reading Relative Risk, Absolute Risk, and NNT, how survival analysis handles competing events that happen first is in When People Don't Stay in Line — The Ones Who Switched, and the Ones Who Left for Another Reason, and how to read signals from non-randomized studies is in What "Adjusted for Age, Sex and BMI" Actually Did — How to Read Confounding and Adjustment. The distance between the numbers a drug lowered and the outcomes a patient gained was covered in Better Numbers, Better Patients? — An Introduction to Surrogate Endpoints, a line drawn over continuous values like a count of five was covered in A Line Painted on a Slope — An Introduction to Reference Ranges and Diagnostic Thresholds, and whether finding more and treating more is always a benefit was asked from the diagnostic side in Is Finding It Early Always Better? — Overdiagnosis. The next installment, The Effect Lives in the Average — The Primer on Applicability of Trial Results, asks whether the average effect of a trial can be applied to the patient in front of me who likely would not have entered the trial, and closes this series.

Source Papers

• "Clinical practice guidelines and quality of care for older patients with multiple comorbid diseases: implications for pay for performance." (2005) — PMID 16091574 — combining guidelines for a hypothetical 79-year-old woman with five conditions yields about 12 medications and $406 per month, most guidelines fail to address multimorbidity, burden, or life expectancy, potentially leading to undesirable outcomes.
• "Clinical guidelines: what happens when people have multiple conditions?" (2018) — PMID 29565938 — perspective paper, combining single-disease guidelines creates interactions and treatment burden, while multimorbidity is omitted from trials and targets misalign.
• "Prevalence of multimorbidity and polypharmacy among adults and older adults: a systematic review." (2024) — PMID 38452787 — 87 studies, typically based on 2 diseases and 5 medications, older adult polypharmacy prevalence ranging from 2.6% to 86.6%, highlighting a need for standardized definition reporting.
• "Polypharmacy: A general review of definitions, descriptions and determinants." (2020) — PMID 31732240 — 5 medications is the most common threshold without consensus, definitions differ by duration, and prevalence ranges from 4% to 57% depending on the definition.
• "Deprescribing in older adults with polypharmacy." (2024) — PMID 38719530 — association of polypharmacy with adverse drug reactions, cognition, function, frailty, falls, hospitalization, and mortality, with just over half of deprescribing trials showing superiority in at least one primary outcome.
• "Reducing inappropriate polypharmacy: the process of deprescribing." (2015) — PMID 25798731 — pill count is the most important predictor of inappropriate prescription and adverse reactions, proposing a 5-step deprescribing process.
• "An update on the clinical consequences of polypharmacy in older adults: a narrative review." (2018) — PMID 30540223 — the association of polypharmacy outcomes is broad, but the authors note that confounding by multimorbidity has been underestimated.
• "Polypharmacy in Older Patients." (2024) — PMID 39475037 — evaluating appropriateness over pill count, conditions for inappropriate prescribing, single-disease guidelines lack stopping criteria and barriers cause medications to accumulate.
• "American Geriatrics Society 2023 updated AGS Beers Criteria® for potentially inappropriate medication use in older adults." (2023) — PMID 37139824 — explicit list of medications older adults should generally avoid, excluding hospice care, intended for the U.S., and not replacing shared decision-making.
• "Treatment burden in multimorbidity: an integrative review." (2024) — PMID 39342121 — components of treatment burden, where increased burden leads to non-adherence, disease progression, reduced quality of life, and caregiver strain.
• "Decision-Making and Downstream Outcomes of the Gabapentinoid-Diuretic Prescribing Cascade." (2025) — PMID 41329486 — records of 120 veterans, gabapentinoids used in differential diagnosis 3.3%, with 23.3% experiencing events likely due to diuretics within 60 days.
• "Reducing prescribing cascades." (2025) — PMID 40171689 — a chain where side effects are misread as new diseases leading to more drugs, giving examples of anticholinergics, calcium channel blockers, and gabapentinoids, along with prevention strategies.
• "A Prescribing Cascade of Proton Pump Inhibitors Following Anticholinergic Medications in Older Adults With Dementia." (2022) — PMID 35814221 — 28,952 older adults with dementia in Nova Scotia, PPI risk ratio 1.38 within 180 days of starting anticholinergic medications, with an incidence of 6.4%.
• "International expert panel's potentially inappropriate prescribing cascades (PIPC) list." (2025) — PMID 40742670 — 12-expert Delphi process, identifying 65 potentially inappropriate prescribing cascades out of 107 candidates.
• "Evaluation of Time to Benefit of Statins for the Primary Prevention of Cardiovascular Events in Adults Aged 50 to 75 Years: A Meta-analysis." (2021) — PMID 33196766 — 8 trials with 65,383 participants, 1 event prevented per 100 people in 2.5 years, per 200 in 1.3 years, per 500 in 0.8 years, with no evidence of mortality benefit.
• "Rationalizing prescribing for older patients with multimorbidity: considering time to benefit." (2013) — PMID 23749475 — definition of time to reach benefit and harm, outcomes must be clearly defined, extraction methods not yet established.
• "The role of lipid lowering medications in the primary prevention of cardiovascular disease and mortality: a meta-analysis of randomized controlled trials." (2026) — PMID 41864911 — 10 pure primary prevention trials with 85,829 participants, all-cause mortality unchanged, major cardiovascular events, myocardial infarction, and stroke reduced.
• "Estimating the Time to Benefit for Preventive Drugs with the Statistical Process Control Method: An Example with Alendronate." (2016) — PMID 26884390 — FIT post-hoc analysis, time to reach benefit 11 months overall, 8 months for those 70 and older, 19 months for those under 70.
• "Statin Therapy for Primary Prevention in Adults Aged 80 Years and Older: A Systematic Review Conducted in Accordance with PRISMA 2020." (2026) — PMID 42706978 — evidence for those 80 and older is indirect and of low certainty, benefit is most plausible in healthy high-risk octogenarians, lacking evidence for routine initiation and requiring individualization.
• "Finding the Balance Between Benefits and Harms When Using Statins for Primary Prevention of Cardiovascular Disease: A Modeling Study." (2019) — PMID 30508425 — model incorporating competing risks, net benefit threshold changing from 14% to 21% in men and 17% to 22% in women (from ages 40–44 to 70–75).
• "The Impact of frailty on the effectiveness of intensive blood pressure control for patients with type 2 diabetes: a secondary analysis of a randomised controlled trial." (2025) — PMID 39915072 — ACCORD BP secondary analysis, benefit of intensive blood pressure control independent of frailty in the standard glycemic group, no benefit in the intensive glycemic group.
• "Safety and benefit of discontinuing statin therapy in the setting of advanced, life-limiting illness: a randomized clinical trial." (2015) — PMID 25798575 — 381 participants, 60-day mortality 23.8% versus 20.3% failing to meet non-inferiority, improved quality of life, authors concluding safety.
• "Discontinuation versus continuation of statins: A systematic review." (2024) — PMID 39051828 — the only RCT shows no difference in 60-day mortality in advanced illness, other non-randomized studies show associations such as a discontinuation-to-mortality risk ratio of 1.92, remaining uncertain.
• "Perceptions of Statin Discontinuation among Patients with Life-Limiting Illness." (2017) — PMID 28520522 — 297 participants, less than 5% worried it would be seen as giving up, 18% told to take them for life, 63% saving money on medication.
• "Effect of Antihypertensive Medication Reduction vs Usual Care on Short-term Blood Pressure Control in Patients With Hypertension Aged 80 Years and Older: The OPTIMISE Randomized Clinical Trial." (2020) — PMID 32453368 — 569 participants aged 80 and older, non-inferior blood pressure control at 12 weeks, systolic blood pressure 3.4 mmHg higher, 66.3% maintained, long-term outcomes unconfirmed.
• "The effect of deprescribing interventions on mortality and health outcomes in older people: An updated systematic review and meta-analysis." (2024) — PMID 39164070 — 259 studies, mortality odds ratio 0.96 in randomized studies (not significant), reduced in the 65–79 age group and patient-tailored subgroups.
• "The feasibility and effect of deprescribing in older adults on mortality and health: a systematic review and meta-analysis." (2016) — PMID 27077231 — non-randomized mortality odds ratio 0.32, randomized 0.82 (not significant), patient-tailored interventions 0.62.
• "Interventions to Address Potentially Inappropriate Prescribing for Older Primary Care Patients: A Systematic Review and Meta-Analysis." (2025) — PMID 40577011 — 118 randomized trials with 417,412 participants, about 0.5 fewer medications, no difference in adverse events, falls, hospitalization, or mortality.
• "Deprescribing for Community-Dwelling Older Adults: a Systematic Review and Meta-analysis." (2020) — PMID 32820421 — comprehensive medication reviews may have reduced mortality (odds ratio 0.74, low certainty), little to no effect on falls, quality of life, or hospitalization.
• "Deprescribing preventive medications in older adults with advanced frailty, dementia, or limited life expectancy: a systematic review and meta-analysis." (2026) — PMID 41840386 — 15 studies, mortality risk ratio 1.15 (not significant, I² 93%), very low certainty, not associated with bad outcomes.
• "Reducing Central Nervous System-Active Medications to Prevent Falls and Injuries Among Older Adults: A Cluster Randomized Clinical Trial." (2024) — PMID 39052289 — 2,367 participants, unable to reduce falls requiring treatment (hazard ratio 1.11).
• "Interventions to improve the appropriate use of polypharmacy for older people." (2023) — PMID 37818791 — Cochrane 3rd update, 38 randomized trials, uncertain whether pharmaceutical care reduces the number of inappropriate medications (very low certainty), little to no difference in hospitalization or quality of life.
• "Polypharmacy Intervention Studies in Older Adults: Challenges in Proving Benefits in Hard Clinical Outcomes." (2025) — PMID 41099738 — limited impact on hard clinical outcomes despite reducing inappropriate drugs, issues with intervention intensity, execution, collaboration, and patient priorities, with multidisciplinary high-intensity interventions showing promise.
• "Outcome preferences of older people with multiple chronic conditions and hypertension: a cross-sectional survey using best-worst scaling." (2019) — PMID 31856842 — 207 participants, stroke is the most feared and treatment burden the least feared, no outcome should be ignored.