A cost-benefit analysis of the life-prolonging effect of nonsteroidal anti-inflammatory drugs for persons with dementia and pain

ABSTRACT

Persons with dementia are more susceptible to pain. If they take Nonsteroidal Anti-Inflammatory Drugs (NSAID) as pain relievers, the policy question is whether these medicines will be socially worthwhile. In this paper, a Cost-Benefit Analysis (CBA) was undertaken of the NSAID to see whether their benefits outweigh the costs for elderly persons with dementia. To construct the benefits, a Quality Adjusted Life Year was used as the intervention output, which was then converted into monetary terms using the Value of a Statistical Life Year. Using estimates based on a large, US national panel data set, the CBA found that the NSAID generated positive net-benefits. It was possible to re-evaluate all the prior dementia interventions that were based on the same data set, using the same methods as for the NSAID CBA. By comparing the NSAID results with those from the other alternative dementia interventions one can establish public funding priorities.

KEYWORDS:

• Pain relievers
• cost-benefit analysis
• dementia
• mortality

JEL CLASSIFICATION:

• I12
• J14

I. Introduction

Both dementia and pain levels are age related. As a consequence, in the community, more than half of the patients with dementia experience daily pain (Barry et al. 2016); and in nursing homes around 60% to 80% of people with dementia regularly experience pain (Corbett et al. 2012). Even when the pain levels were the same as for those non-cognitively impaired, the pain levels of those with dementia has been found to be more pronounced (Defrin et al. 2015). With 6.7 million Americans in 2023 living with Alzheimer’s dementia (Alzheimer’s Association 2023), which is just the main type of dementia, there are many persons who would want to know if there exists anything that is effective to treat this dementia associated pain.

There are non-drug interventions that are effective in reducing pain. The most-promising ones are exercise and/or movement therapies (e.g. rocking chair), massage and/or human touch, heat and relaxation therapies, and human interaction and presence. The majority of randomized clinical trials (RCTs) of non-drug pain interventions have excluded persons with dementia (Achterberg et al. 2020). Thus, these studies do not provide the necessary information to guide interventions for alleviating dementia associated pain.

As for drug medicines for reducing pain (analgesics) for those with dementia, there are two categories: opioid and non-opioid. Opioids are for moderate to severe pain. The problem with using opioids is that they increase the risk of dementia (the hazard ratio is 1.86 for those who use opioids relative to those who do not use opioids, Sun et al. 2023). That leaves non-opioids as the main viable category of analgesic in those with dementia with chronic pain.

The most important non-opioid medications are the non-steroidal anti-inflammatory drugs (NSAID), the principal ones being aspirin (Bayer), ibuprofen (Advil) and Naproxen (Aleve). It is the NSAID that will be the main focus of the analysis in this paper. However, acetaminophen (Tylenol) will also be covered which, although not an anti-inflammatory drug, it will be able to be tested whether the medical literature is correct to regard this medication to be the mainstay for treating mild-to-moderate pain in advanced dementia (Achterberg et al. 2020).

The objective of this paper is to carry out a Cost-Benefit Analysis (CBA) of the simple analgesics, the NSAID, for persons with dementia and pain. The evaluation will be carried out using a large, US national data set provided by the National Alzheimer’s Coordinating Center (NACC). The main outcome that is going to be used to make the evaluation is the mortality changes from the medications. This mortality outcome has been used for a number of CBAs of dementia interventions, see Brent (2022). Its application here makes it possible to compare the relative desirability of the NSAID to other evaluated interventions using the NACC data set, in order to inform policy makers when deciding which dementia interventions to fund.

In the next section, the methodology that will be used to value the benefits of the NSAID is explained. Section III presents the mortality estimation framework which covers the regression equation and a specification of all the controls. Section IV covers the data source and provides measures of mortality and the main independent variables. This is followed in section V by the definitions of all the variables that are in the data set used in the study, which leads to the data summary.

Section VI has the estimation results, and section VII uses these results to carry out the CBA. The sensitivity analysis is in section VIII and this involves providing an alternative estimation technique from the regression analysis that is used to provide the main results. The discussion in section IX exploits the fact that many of the controls in the mortality estimation equation for the NSAID are also interventions that have previously been evaluated by CBA using the same data set. This enables a direct comparison of the CBA results for the NSAID to be made with the previously evaluated interventions for policy purposes, as a common methodology can now be applied. The summary and conclusions are in the final section of the paper.

II. The method for valuing the benefits

The effects of the NSAID will be given by their impacts on a person’s Quality Adjusted Life Years (QALYs), which is the product of one’s life years (LY) and the quality of each life year (QoL). The QALY is a comprehensive measure of effects, as any intervention that takes place affects either a person’s quantity or quality of life, or both. In our evaluation of the analgesics, the QoL will be assumed to be fixed, $\overline{\mathrm{QoL}}$, as there was no evidence in our data that it was affected by any of the NSAID analysed. The life years that will be affected by the analgesics is by the medications altering mortality through the probability of dying π, and thereby not experiencing the remaining life years YR that a person has. Thus, expected life years, LY, are obtained by not dying and LY = (1 − π)YR.

The benefits B of the effects of the NSAID are defined as the value of any change in QALYs generated, where value will be determined by the Value of a Statistical Life Year (VSLY). The VSLY is derived from a person’s labour risk of dying-wage trade-off, which is considered to be fixed at $\overline{\mathrm{VSLY}}$in our analysis and determined by the literature.¹

Thus, the monetary value of any QALYs is given as:

(1) $\mathit{VSLY}\mathit{\hspace{0.17em}}\times \mathit{\hspace{0.17em}}\mathit{QALYs}=\mathit{VSLY}\mathit{\hspace{0.17em}}\times \mathit{\hspace{0.17em}}\left[\mathit{QoL}\mathit{\hspace{0.17em}}\times \mathit{\hspace{0.17em}}\mathit{LY}\right]=\overline{\mathit{VSLY}}\times \mathit{\hspace{0.17em}}[\overline{\mathit{QoL}}\times \mathit{\hspace{0.17em}}(1-\mathit{\pi }\mathit{\hspace{0.17em}})\mathit{YR}]$(1)

and the monetary value from the change in QALYs from the NSAID are:

(2) $\mathrm{B}=\mathit{\hspace{0.17em}}\overline{\mathit{VSLY}}\frac{\mathrm{d}\left(\mathrm{QALYs}\right)\mathit{\hspace{0.17em}}}{\mathit{dNSAID}}=\overline{\mathrm{VSLY}}\mathit{\hspace{0.17em}}\times \mathit{\hspace{0.17em}}\overline{\mathrm{QoL}}\left[-\mathit{\hspace{0.17em}}\frac{\mathrm{d\pi }}{\mathrm{dNSAID}}\right]\mathrm{YR}=\overline{\mathrm{VSLY}}\mathit{\hspace{0.17em}}\times \mathit{\hspace{0.17em}}\overline{\mathrm{QoL}}\left[-{\mathit{\alpha }}_1\right]\mathrm{YR}$(2)

where: $\frac{\mathrm{d}\mathit{\pi }}{\mathrm{dNSAID}}={\mathit{\alpha }}_1$.

It is the effect of the NSAID on the probability of dying (mortality), α₁ , that will therefore determine whether the benefits are positive or negative. Positive benefits require that NSAID lower the risk of dying, with α₁ <0. How α₁ is to be estimated will now be explained.

III. Mortality estimation framework

The regression equation

Since panel data will be used for the estimation, all variables will have the subscripts ij, where i denotes the client involved, and j stands for the visit number.² The dependent variable in the regression is the probability of dying π. The main independent variables are the NSAID and the level of dementia D. Dementia needs to be included, as it is those with dementia that the pain relievers are targeting, and dementia on its own will reduce a person’s life expectancy.

In addition to the main independent variables, there will be three sets of controls represented by X. X_N are the controls related to the newly evaluated, non-pharmaceutical dementia interventions that have been shown by a CBA to be socially worthwhile. X_P are the controls that represent recently evaluated pharmaceutical interventions (other than the NSAID) for people with dementia. The third set are health controls X_H specified in the health economics literature to be the main determinants of the mortality of older adults. Therefore, the mortality regression is (using a panel data, two-way, random effects construction):

(3) ${\pi }_{\mathit{iv}}={\alpha }_0+{\alpha }_1\mathit{NSAI}{D}_{\mathit{iv}}+{\alpha }_2{D}_{\mathit{iv}}+{\alpha }_j{X}_{\mathit{iv}}+{\alpha }_{\mathit{i}}+{\alpha }_{\mathit{v}}+{\epsilon }_{\mathit{iv}}$(3)

where the α coefficients are the regression parameters to be estimated, α_i and α_v are the client- and visit-specific dummy variables that are treated as error terms, and ε_iv is the overall random error term. α₁ is the coefficient of interest as defined in Equation (2)(2) $\mathrm{B}=\mathit{\hspace{0.17em}}\overline{\mathit{VSLY}}\frac{\mathrm{d}\left(\mathrm{QALYs}\right)\mathit{\hspace{0.17em}}}{\mathit{dNSAID}}=\overline{\mathrm{VSLY}}\mathit{\hspace{0.17em}}\times \mathit{\hspace{0.17em}}\overline{\mathrm{QoL}}\left[-\mathit{\hspace{0.17em}}\frac{\mathrm{d\pi }}{\mathrm{dNSAID}}\right]\mathrm{YR}=\overline{\mathrm{VSLY}}\mathit{\hspace{0.17em}}\times \mathit{\hspace{0.17em}}\overline{\mathrm{QoL}}\left[-{\mathit{\alpha }}_1\right]\mathrm{YR}$(2) .³

The selection of the controls

Non-Pharmaceutical controls X_N

A number of non-pharmaceutical interventions have been newly evaluated and found to be socially worthwhile. These were vision correction, Medicare eligibility, and avoiding living in a nursing home (since living in a nursing home increases mortality), Brent (2022). As these interventions were estimated using a causal strategy, they are exogenous and can be used as controls for the main variables. Their role as controls is that the effect of the NSAID must be estimated to be over and above these successful existing interventions

Pharmaceutical controls X_P

Prior to this CBA of the NSAID, a number of other pharmaceutical interventions were evaluated and found to increase mortality and thus not to be socially worthwhile. These were FDA-approved medications for dementia (Brent 2023), beta-blockers (Brent 2024a), and antipsychotics (2024b). Again, they were estimated using causal strategies, so are exogenous and also can be used as controls.

Health controls X_H

More than 75% of the gains in life expectancy today are realized after 65 years of age (Eggleston and Fuchs 2012). Our sample of older adults with their behaviour and characteristics is therefore highly relevant for examining the possible determinants of mortality. Variables will be assigned to categories that have been identified by the health economics literature to be some of the main determinants of adult mortality (see for example Cutler, Deaton, and Lleras-Muney 2006; Hummer, Rogers, and Eberstein 1998; Shaw, Horrace, and Vogel 2005). For the demographic variables, there is age, race and gender; for nutrition there is height and birth month; and for medical reasons there is BMI, smoking behaviour and the QoL. One weakness of the data set is that it does not have a public health variable, such as types of immunization. But, as an environmental or locality proxy, the type of residence housing a client lives in, community or independent living, can be used, seeing that the type of housing can affect mortality outcomes if illnesses are contagious.

IV. The data source, measures of mortality and the main independent variables

Data source

The data that will be used to estimate the effect of NSAID on mortality come from the National Alzheimer’s Coordinating Center (NACC). NACC has constructed a panel data set that has been operational since 2005, called the Uniform Data Set (UDS). These data consist of demographic, clinical, diagnostic, and neuropsychological information on participants with normal cognition, mild cognitive impairment, and dementia who visited 32 US Alzheimer’s Disease Centers (ADC). In our analysis for this study, there were 33,794 visits recorded, for 16,526 individuals, covering up to 12 visits per client, over a thirteen-year period. This data set is fully explained elsewhere (Beekly et al. 2007; Morris et al. 2006; Weintraub et al. 2009). The UDS was also the data source used for all the prior CBAs of interventions covered in this paper.

Mortality

A person is judged to have died if the person is known to be deceased. The person is classed not dead if the person is not deceased, or is unknown to be deceased.

Dementia

The instrument that will be used to measure dementia is the Clinical Dementia Rating (CDR) scale, known as the CDR® Dementia Staging Instrument, created by Washington University. The CDR is based primarily on a neurological exam and informant reporting, see Morris (1997). A CDR was administered to each NACC participant at each visit by a clinician. There are six domains in the CDR: memory, orientation, judgement and problem solving, community affairs, home and hobbies, and personal care. Each domain is assessed using a 0 to 3 interval (none, mild, moderate and severe) with a questionable response being scored as 0.5. The CDR-SB (the CDR sum of boxes) is the aggregate score across all six domains and this has a range of 0 to 18. As the focus is pain relievers for those with dementia, our sample ensures that all the persons analysed had some degree of dementia (the CDR was 0.5 or higher). This excluded 41% of the NACC visits.

The nonsteroidal anti-inflammatory drugs

Medications included in this NSAID category are: non-steroidal anti-inflammatory agents, salicylates, COX2 inhibitors, and analgesic combinations. There were 56 different medications listed under this broad category of medicines in the NACC data set.⁴ Of those who ever reported taking one medication at any visit, the main ones were, aspirin, ibuprofen and naproxen (their frequencies are given in Table 2). In the estimation results section, the estimates for the NSAID category as a whole will be presented, together with the three main NSAIDs separately to show their individual impact. As explained earlier, the results separately for acetaminophen will also be shown because of its emphasis as the main pain reliever in the medical literature for those with dementia.

V. The definitions of all the variables used in the study and the data summary

All the variables used in the mortality estimation equation are listed in Table 1, together with their definitions. The definitions come from NACC’s ‘Description of NACC Derived Variables to be Used in Data Analysis’ (August 2014) and NACC’s Uniform Data Set (UDS) ‘Coding Guidebook for Initial Visit Packet’ (last modified 14 January 2014).

Table 1. Definitions of all the variables.

Variable	Description
Mortality π	Subject is known to be deceased. 1 = Deceased; 0 = not deceased, or unknown to be deceased.
Dementia D	Clinical Dementia Rating (CDR) Sum of Boxes (SB). Total CDR score based on Memory, Orientation, Judgement & Problem Solving, Community Affairs, Home & Hobbies, Personal Care, each of the six categories on a scale of 0 to 3.
NSAID	Reported current use of nonsteroidal anti-inflammatory medication. 1= Reported use at visit; 0 = Did not report use at visit.
Aspirin	Reported use of Aspirin. 1 = Reported use of Aspirin at any visit; 0 = Did not report use at any visit.
Ibuprofen	Reported use of Ibuprofen. 1 = Reported use of Ibuprofen at any visit; 0 = Did not report use at any visit.
Naproxen	Reported use of Naproxen. 1 = Reported use of Naproxen at any visit; 0 = Did not report use at any visit.
Acetaminophen	Reported use of Acetaminophen. 1 = Reported use of Acetaminophen at any visit; 0 = Did not report use at any visit.
Corrective lenses	If the subject usually wears corrective lenses, is the subject’s vision functionally normal with corrective lenses? 1 = Yes; 0 = No, if any functional impairment exists (reduced ability to do everyday activities such as reading, watching television).
Medicare eligibility	Is the subject eligible for Medicare? Age ≥ 65 = 1; Age < 65 = 0.
Nursing Homes	Is the subject’s type of residence a skilled nursing facility/nursing home? 1 = Yes; 0 = No.
Antipsychotics	Reported current use of an antipsychotic agent. 1= Reported use at visit; 0 = Did not report use at visit.
Beta-Blockers	Reported current use of a beta-adrenergic blocking agent. 1= Reported use at visit; 0 = Did not report use at visit.
FDA-Approved Medications	Reported current use of an FDA-approved medication for Alzheimer’s disease symptoms. 1= Reported use at visit; 0 = Did not report use at visit.
Age	Subjects age at time of visit.
White	NIH race definition. 1 = White; 0 = Non-White.
Female	Subject’s sex. Female = 1; Male = 0.
Height	Subject’s height in inches.
Birth month	Month the person was born. 1 = Born between October and December; 0 = Not born between October and December.
BMI	Body Mass Index. The standardized calculation used is: BMI = weight (lbs) ÷ [height (in)]^2 × 703.
Smoker years	Total number of years smoked cigarettes.
QoL	1 - [Total Geriatric Depression Scale (GDS) score/15].
Community living	Does the subject reside in a retirement community or independent group living? 1 = Yes; 0 = No.
Assisted living	Does the subject reside in assisted living/boarding home/adult family home? 1 = Yes; 0 = No.

It is important to understand that the three individual NSAID are defined slightly differently from the group NSAID, as they record the use of a respective medication at any visit a person made to the ADC clinic while for the NSAID, the data record their use at just the last visit to a clinic. This means that the individual NSAID are not simple decompositions of the group NSAID data. Though, the individual NSAID do give a good indication of what medications the group NSAID was most likely to contain.

Table 2 gives the data summary in terms of the number of observations, mean values, standard deviations, and minimum and maximum values. 26% of the clients died at some time at which their visits were recorded. As many as 44% of the clients took the NSAID at the current visit. Of those taking a NSAID at any visit, about one-sixth used aspirin (18%). Much smaller shares were ibuprofen (3%) and naproxen (4%). Acetaminophen had 4% of users at any visit. The dementia score was 3.7, signifying that, on average, the clients had a score that exceeded a complete span for one of the six CDR categories.

Table 2. Descriptive statistics for all the variables.

Variable	Number	Mean	Standard Deviation	Minimum	Maximum
Mortality π	33,794	0.2599	0.4386	0	1
Dementia	33,794	3.7448	3.4847	0.5	18
NSAID	33,794	0.4374	0.4961	0	1
Aspirin	32,816	0.1769	0.3816	0	1
Ibuprofen	32,816	0.0271	0.1624	0	1
Naproxen	32,816	0.0352	0.1843	0	1
Acetaminophen	32,816	0.0378	0.1908	0	1
Corrective lenses	33,794	0.9430	0.2318	0	1
Medicare eligibility	33,794	0.8682	0.3282	0	1
Nursing homes	33,794	0.0061	0.0777	0	1
Antipsychotics	33,794	0.0445	0.2063	0	1
Beta-Blockers	33,794	0.2331	0.4228	0	1
FDA-Approved Medications	33,794	0.4676	0.4990	0	1
Age	33,794	75.4171	9.5714	21	107
White	33,794	0.8384	0.3681	0	1
Female	33,794	0.4954	0.5000	0	1
Height	33,794	65.7053	4.1536	37	82
Birth month	33,794	0.2512	0.4337	0	1
BMI	33,794	26.7881	4.9466	12.1	84.2
Smoker years	33,794	10.9138	15.7073	0	82
QoL	33,794	0.8289	0.1816	0	1
Community Living	33,794	0.0682	0.2522	0	1
Assisted Living	33,794	0.0304	0.1718	0	1

Representing the non-pharmaceutical controls: 94% of those who wore corrective lenses had vision that was functionally normal; 87% were Medicare eligible; and around 1% lived in nursing homes that potentially could be avoided. Representing the non-pharmaceutical controls: 4% of the sample took antipsychotics; 23% took beta-blockers, and 47% used FDA-approved dementia medications. As for the main health controls: the average age was 75 years; 84% of the clients were white; and exactly half were females.

VI. Estimation results

The main estimate of the effect of the NSAID group on mortality, given by α₁ in Equation (3)(3) ${\pi }_{\mathit{iv}}={\alpha }_0+{\alpha }_1\mathit{NSAI}{D}_{\mathit{iv}}+{\alpha }_2{D}_{\mathit{iv}}+{\alpha }_j{X}_{\mathit{iv}}+{\alpha }_{\mathit{i}}+{\alpha }_{\mathit{v}}+{\epsilon }_{\mathit{iv}}$(3) , is in column (1) in Table 3. In columns (2) to (4) are the alternative mortality effect estimates when just an individual NSAID is highlighted. Column (5) has the acetaminophen effect estimate. All columns include also the estimates of the three sets of controls.

Table 3. Marginal effects estimates for random effects models of pain relievers on mortality using logit (p-values in parentheses).^†

Variables	(1)	(2)	(3)	(4)	(5)
Main Variables
Dementia	0.0119*** (0.000)	0.0183*** (0.000)	0.0021*** (0.000)	0.0129*** (0.000)	0.0152*** (0.001)
NSAID α1	−0.0135*** (0.000)
Aspirin		−0.0214*** (0.003)
Ibuprofen			−0.0042** (0.031)
Naproxen				−0.0028 (0.777)
Acetaminophen					0.0084 (0.255)
Non-pharmaceutical Controls
Vision correction	−0.0161*** (0.007)	−0.0307*** (0.011)	−0.0055*** (0.001)	−0.0120* (0.071)	−0.0218*** (0.008)
Medicare eligibility	−0.0713*** (0.000)	−0.0950*** (0.000)	−0.0151*** (0.000)	−0.0704*** (0.000)	−0.0826*** (0.000)
Nursing Homes	0.0738** (0.040)	0.1243*** (0.000)	0.0303*** (0.000)	0.0891*** (0.005)	0.1090*** (0.000)
Pharmaceutical Controls
Antipsychotics	0.0154* (0.068)	0.0281** (0.042)	0.0040** (0.051)	0.0230** (0.051)	0.0150 (0.171)
Beta Blockers	0.0119*** (0.005)	0.0178** (0.038)	0.0013 (0.123)	0.0132*** (0.007)	0.0137** (0.023)
FDA-Approved Medications	0.0210*** (0.000)	0.0217*** (0.009)	0.0020** (0.024)	0.0278*** (0.000)	0.0276** (0.023)
Health Controls
Age	0.0082*** (0.000)	0.0112*** (0.000)	0.0012*** (0.000)	0.0091*** (0.000)	0.0101*** (0.000)
White	0.0938*** (0.000)	0.1091*** (0.000)	0.0087*** (0.000)	0.1083*** (0.000)	0.1014*** (0.000)
Female	−0.0553*** (0.000)	−0.0623*** (0.000)	−0.0071*** (0.000)	−0.0492*** (0.000)	−0.0625*** (0.000)
Height	0.0016** (0.036)	0.0022 (0.171)	0.0032** (0.034)	0.0021* (0.058)	0.0014 (0.189)
Birth month	−0.0085 (0.150)	−0.0033 (0.782)	−0.0016 (0.108)	−0.0023 (0.732)	−0.0017 (0.832)
BMI	−0.0025*** (0.000)	−0.0032*** (0.001)	−0.0003*** (0.001)	−0.0021*** (0.002)	−0.0027*** (0.000)
Smoker	0.0009*** (0.000)	0.0010*** (0.001)	0.0001*** (0.001)	0.0011*** (0.000)	0.0011*** (0.000)
QOL	−0.0390*** (0.000)	−0.0479** (0.033)	−0.0088*** (0.000)	−0.0514*** (0.000)	−0.0536*** (0.000)
Community Living	0.0418*** (0.000)	0.0584*** (0.000)	0.0091*** (0.000)	0.0521*** (0.000)	0.0526*** (0.000)
Assisted Living	0.0583*** (0.000)	0.0906*** (0.000)	0.0272*** (0.000)	0.0698*** (0.000)	0.0826*** (0.000)
Number of Obs.	33,794	13,624	29,636	27,070	20,470
Number of Individuals	16,526	7,606	14,844	13,734	10,835
Chi-Square^††	16,000*** (0.000)	5,472*** (0.000)	13,000*** (0.000)	12,000*** (0.000)	8,758*** (0.000)

^† Significance levels on coefficients: * 10%; ** 5%; *** 1%.

^††The Chi-square is for a likelihood-ratio test that ρ = 0, which implies that there would be no difference between using a pooled sample rather than a panel for estimation.

^†††All equations also include 11 of the 12 visit numbers.

As expected, dementia increased the chances of dying in all the columns, and this coefficient was always highly significant at least the 1% level. In column (1), all the controls were significant at least the 5% level, and had the expected signs, except for the antipsychotic medications and the birth month, which were insignificant. The main estimate of α₁ is − 0.0135, which means that the NSAID reduced mortality. Because of the large monetary value that is to be assigned to any mortality reduction, the benefits of the NSAID will be economically significant.

From the individual estimates of the NSAID on mortality in columns (2) to (4), it can be seen that the dominant contributor to the large impact of the NSAID group is due to Aspirin. Aspirin has the largest impact at − 0.0214 and this is highly statistically significant. Also statistically significant is Ibuprofen, though its contribution is much smaller at − 0.0042. Naproxen was not a significant contributor. Note that Acetaminophen in column (5) was not significant, contrary to the medical literature’s emphasis of it as the main analgesic.

VII. The cost-benefit analysis

In the Cost-Effectiveness Analysis of Aspirin by Greving et al. (2008), which was the dominant NSAID in our data set, age 55 was the starting date for the evaluation and 10 years was the assumed period that Aspirin had to be taken in order to obtain the effects estimated. The same time line will be assumed for our CBA of the NSAID. This means that the annual costs will be projected over 10 years, and the mortality reduction will be applied to 20 remaining years of life (the difference between the average age of 75 years in our data set and the assumed 55 year starting date).

The costs

Since the NSAID are over-the-counter medicines, they did not require a physician to write a prescription for these medicines. Therefore, there are no other costs to include in the CBA than just the payments for the medicines themselves. On the basis of a national data set of wholesale prices for the NSAID, for maximum daily prescribed doses (and therefore a conservative estimate), the average cost per patient of a 30-day supply for NSAID was 104.93 USD (Rasu et al. 2014), making the annual cost 1,259 USD in 2000–2007 prices. Over 10 years, the cost would be 12,590 USD. Discounting the 10 years of annual costs at the same 3% rate as used for all the dementia intervention evaluations, leads to a present value cost for the NSAID of 11,063 USD per person.

The benefits

From Equation (2)(2) $\mathrm{B}=\mathit{\hspace{0.17em}}\overline{\mathit{VSLY}}\frac{\mathrm{d}\left(\mathrm{QALYs}\right)\mathit{\hspace{0.17em}}}{\mathit{dNSAID}}=\overline{\mathrm{VSLY}}\mathit{\hspace{0.17em}}\times \mathit{\hspace{0.17em}}\overline{\mathrm{QoL}}\left[-\mathit{\hspace{0.17em}}\frac{\mathrm{d\pi }}{\mathrm{dNSAID}}\right]\mathrm{YR}=\overline{\mathrm{VSLY}}\mathit{\hspace{0.17em}}\times \mathit{\hspace{0.17em}}\overline{\mathrm{QoL}}\left[-{\mathit{\alpha }}_1\right]\mathrm{YR}$(2) , it can be seen that there are four ingredients that make up the estimate of the benefits, that is: $\overline{\mathrm{VSLY}},\overline{\mathrm{QoL},}{\alpha }_1\mathrm{and}\mathrm{YR}$. With YR = 20, the amount of LYs gained from the mortality effect of the NSAID (given in Table 3 as ${\alpha }_1$ = 0.0135) was 0.27 years. Table 2 reports that the average QoL of a LY in our sample was 0.83. This makes the QoL adjusted gain of LYs, that is, the number of QALYs, equal to 0.224. Using the VSL from Brent (2023), and using the time line used by Greving et al. (2008), one obtains $\overline{\mathrm{VSLY}}$ = 163,500 USD.⁵ Valuing the gain in QALYs by this $\overline{\mathit{VSLY}}$ makes the benefits equal to 36,624 USD per person.

The net-benefits

With the benefits equal to 36,624 USD and the costs 11,063 USD, the net-benefits were highly positive equal to 25,561 USD per person, with a benefit-cost ratio over 3. Since the baseline for the net-benefits is year 2000, the net-benefits in 2024 prices would be 46,636 USD per person.⁶ The NSAID was socially worthwhile unlike previously evaluated medications for the elderly with dementia.

VIII. Sensitivity analysis

To test whether the main estimates of the pain-relievers obtained from the random effects model were robust, an alternative estimation technique will be applied, that is, Propensity Score Matching (PSM). PSM is an approach that attempts to replicate the classical randomized clinical trial (RCT) methodology, see Austin (2011). It does this by seeking to obtain the average treatment effect (ATE), the difference between the outcomes for those in the treatment group, who take the NSAID, and those in the control group, who do not take any pain relievers, by creating a new control group. This new control group is obtained by matching all the characteristics of the treatment group except for the taking of the NSAIDs (using Probit). Thus, the only reason that the ATE exists is due to the NSAID being taken and not something else concerning the two groups. The PSM estimates of the ATEs are in Table 4.

Table 4. Average treatment effects of pain relievers on mortality using logit for propensity score matching (p-values in parentheses).^†

Variables	(1)	(2)	(3)	(4)	(5)
NSAID: α1	−0.0455*** (0.000)
Aspirin		−0.0380*** (0.000)
Ibuprofen			−0.0579*** (0.000)
Naproxen				−0.0252 (0.140)
Acetaminophen					0.0054 (0.711)
Obs. Raw Obs. Matched	33,794 67,588	13,624 27,248	29,636 59,272	27,070 54,140	20,470 40,940

^† Significance levels on coefficients: *10%; **5%; ***1%.

Overall, the estimates resulting from using PSM confirm the main findings from using regression analysis. The NSAID produced a statistically significant reduction in mortality. In fact, the PSM estimate was much more impactful than that given by the main estimates, as α₁ was over twice the size. One can therefore be confident that the main results give a conservative estimate of the benefits of the NSAID.

As with the main results, it was Aspirin and Ibuprofen that were the main contributors to the NSAID effect. The difference this time was that Ibuprofen’s contribution was now higher that for Aspirin. It was still the case that Naproxen and Acetaminophen made no contribution.

IX. Discussion

A number of the controls in the mortality regression are interventions for persons with dementia. Because of this dual role for some of the controls as interventions, and interventions as controls, one is able to provide here a more comprehensive re-evaluation of many of the previously published dementia interventions, using the same benefit methodology that was used in this paper for the NSAID. In the process, one obtains CBA results that enable meaningful comparisons to be made about the relative desirability of a number of dementia interventions, to see whether the NSAID should be a priority for policymakers deciding how to allocate scarce health care funds.

There will be seven interventions for which CBA results will be generated in this section: three non-pharmaceutical interventions, three pharmaceutical interventions, and the NSAID intervention. As explained earlier, the three non-pharmacological interventions are vision correction, Medicare eligibility, and avoiding nursing homes. The three pharmaceutical interventions are the antipsychotics, the beta-blockers, and the FDA-approved dementia medications. The NSAID intervention evaluation results are repeated here for ease of comparison with the other interventions. Table 5 shows the CBA results in USD expressed in 2000 prices for all the interventions.

Table 5. Cost-benefit analysis of six interventions compared with NSAID.

Variables	Mortality Rate	QALYs Changed	Benefits	Costs	Net-Benefits
NSAID	˗ 0.0135	0.2241	36,640	11,063	25,577
Antipsychotics	0.0154	˗ 0.2556	˗ 41797	62,852	˗ 104,649
Beta-Blockers	0.0119	˗ 0.1975	˗ 32298	797	˗ 33095
FDA-Approved Medications	0.0210	˗ 0.3486	˗ 56996	31,949	˗ 88945
Vision Correction	˗ 0.0161	0.2673	43,697	765	42,932
Medicare Eligibility	˗ 0.0713	1.1836	193,515	17,444	176,071
Avoiding Nursing Homes	˗ 0.0738	1.2251	200,301	54,245	146,056

We now explain how Table 5 was constructed and what the results imply for policy-makers.

Constructing Table 5

The logic of using year 2000 as the common base year for the monetary values for all the evaluations is that one cannot expect that any of the interventions would produce instantaneous benefits. The interventions need to be in place for a large number of years (to be conservative, say 20 years). No one taking the NSAID today would see their mortality rate decline today. Similarly, for beta-blockers; they could not have been found to be so adverse unless they were used persistently. If people had stopped taking the beta-blockers 20 years ago, they would have avoided the adverse life expectancy decline today. The mortality rate changes shown in Table 5 were derived from the mortality estimates in Table 3.

The QALYs gained for the NSAID were estimated using $\overline{\mathrm{QoL}}\left[-{\mathit{\alpha }}_1\right]\mathrm{YR}$ from Equation (2)(2) $\mathrm{B}=\mathit{\hspace{0.17em}}\overline{\mathit{VSLY}}\frac{\mathrm{d}\left(\mathrm{QALYs}\right)\mathit{\hspace{0.17em}}}{\mathit{dNSAID}}=\overline{\mathrm{VSLY}}\mathit{\hspace{0.17em}}\times \mathit{\hspace{0.17em}}\overline{\mathrm{QoL}}\left[-\mathit{\hspace{0.17em}}\frac{\mathrm{d\pi }}{\mathrm{dNSAID}}\right]\mathrm{YR}=\overline{\mathrm{VSLY}}\mathit{\hspace{0.17em}}\times \mathit{\hspace{0.17em}}\overline{\mathrm{QoL}}\left[-{\mathit{\alpha }}_1\right]\mathrm{YR}$(2) , where the estimate of the mortality decline was equal to 0.0135. In the calculation of the NSAID benefits, $\overline{\mathrm{QoL}}$ was set to 0.83 and YR = 20. As the LY gained was 20 times the mortality rate, the gain was 0.27 years. Multiplying this gain by the QoL produced the 0.2241 QALY figure that is in Table 5. Applying the same method to the other interventions, that is, multiplying the particular mortality rates changes by the $\overline{\mathrm{QoL}}$ of 0.83 and YR of 20, leads to the number of QALYs changed in Table 5.

Valuing the QALYs changed for each of the interventions by the VSLY of 163,500 USD produces the benefits estimates. The costs for the NSAID were estimated earlier in this paper. The costs for the other interventions were taken from the CBAs in the literature. Note that all the costs were expressed in 2000 USD and so were on a comparable basis with the benefits.

The net-benefits are the differences between the benefits and the costs. They are positive for the NSAID and the non-pharmaceutical interventions, and negative for the pharmaceutical interventions.⁷

The policy implications of the CBA results

It was the size of the mortality reduction for the NSAID and the non-pharmaceutical interventions that mainly made their net-benefits positive. If there were no budget constraint for the person or entity financing any dementia intervention, then the NSAID and non-pharmaceutical interventions would all be very socially worthwhile and justify funding. Since the benefit-cost ratio for the NSAID was lower than for the three non-pharmaceutical interventions,⁸ if there were a budget constraint, the size of the budget constraint for dementia would determine whether the NSAID would be approved. Since these four interventions are not mutually exclusive, to finance all four of the worthwhile dementia interventions, the budget would have to total 83,517 USD per person. This would generate 390,636 USD per person in net-benefits.

The net-benefits for the three pharmaceuticals had to be negative, given that they increased mortality, making the benefits negative. There would therefore be no positive benefits to offset the negative costs. Nonetheless, it can still be useful to calculate the actual net-benefits of the pharmaceuticals to see exactly how much they decreased social welfare. With this knowledge, one can see how the pharmaceutical interventions compare to the NSAID, if the pharmaceuticals were viewed as being interventions being avoided, and thus prevented from having taken place by, say, a nursing home ombudsman.

Thus, if the pharmaceutical interventions did not take place in the past, and were now being reversed, such that people would be allowed to avoid taking them, then this would generate mortality benefits and the costs would effectively be reimbursed. The negative net-benefit figures for the three pharmaceuticals in Table 5 would become positive, and actually exceed those of the NSAID and give them priority for funding. In fact, if there were a budget constraint, the pharmaceutical intervention reversals would be given priority over any of the other interventions, because the pharmaceuticals would generate only cost savings, and thus require no funding at all, making any budget constraint for them irrelevant.⁹

X. Summary and conclusions

People with dementia experience pain intensively. Pain relievers do exist in the form of NSAID. The main issue is whether those taking the pain relievers do so at the expense of reducing their life expectancy. In this paper a CBA is undertaken to see if social welfare is increased or decreased by the NSAID when all the benefits and costs are included. Using a large, US national data set, the main finding was that these pain relievers did in fact increase social welfare. This finding was the case even when in the sensitivity analysis an alternative estimation technique was employed.

To aid identification for the estimations, which focused on the effect of the NSAID on mortality, a large number of controls were employed. The three sets of controls consisted of non-pharmaceutical, pharmaceutical and health variables. The non-pharmaceutical and pharmaceutical controls selected were interventions in their own right, and were subject to previous CBA evaluations in the literature. Because these interventions were included as controls in the regression used to estimate the NSAID impact on mortality, their independent effect on mortality was determined. This enabled all the previous interventions to be subject to a new CBA evaluation using the same methods as was used in this paper to carry out the CBA of the NSAID. In this way, all the previous evaluations could be directly compared with the NSAID CBA to see how public policy priorities could be determined as to how best to spend public funds for those with dementia.

Of the six interventions that were compared with the NSAID intervention, the three non-pharmaceutical interventions were confirmed to be socially worthwhile, and the three pharmaceutical interventions were confirmed to be not socially worthwhile. As the NSAID intervention was also found to be socially worthwhile in this paper, this meant that, for the first time, a medicine can be added to the list of interventions that deserve public funding.

Because a full CBA was carried out of the three pharmaceutical interventions, that were not the NSAID, the exact amount of their negative net-benefits was ascertained. This was important in connection with the fact that medicines were interventions that could be reversed, unlike, say, medical operations. If medications that are harmful cease to be taken, then this is a win-win situation. The harm that now ceases becomes a positive benefit, and the expenditure on the medicines that now is avoided becomes a cost-saving. As a result, what previously was a negative net-benefit becomes a precise positive net-benefit amount. When the three pharmaceutical interventions were viewed as medications that were no longer taking place, no costs were involved. Therefore, these reverse interventions would be given priority whenever a budget constraint existed for a public policymaker.

Notwithstanding the contributions made in this paper, for completeness, one needs to acknowledge the weaknesses in our study. It was not known exactly which particular NSAID contributed to the group NSAID reduction in mortality; although it probably was Aspirin and Ibuprofen. Nor was it known the duration that the medications had been taken for, and what was the dosage. Thus, some simplifying assumptions were required to carry out the CBA of the NSAID. In addition, because no measure of pain assessment existed in the NACC data set, the paper did not actually show that pain was reduced, and by how much it was reduced. Instead, it had to be implied that because mortality was reduced, it was likely that pain had been reduced. Hadjistavropoulos et al. (2014) does refer to existing pain assessment measures, such as videoing facial expressions, and it would have been useful as a background to the CBA if such measures did exist in the data set.

Acknowledgements

The NACC database is funded by NIA/NIH Grant U24 AG072122. NACC data are contributed by the NIA-funded ADRCs: P30 AG062429 (PI James Brewer, MD, PhD), P30 AG066468 (PI Oscar Lopez, MD), P30 AG062421 (PI Bradley Hyman, MD, PhD), P30 AG066509 (PI Thomas Grabowski, MD), P30 AG066514 (PI Mary Sano, PhD), P30 AG066530 (PI Helena Chui, MD), P30 AG066507 (PI Marilyn Albert, PhD), P30 AG066444 (PI John Morris, MD), P30 AG066518 (PI Jeffrey Kaye, MD), P30 AG066512 (PI Thomas Wisniewski, MD), P30 AG066462 (PI Scott Small, MD), P30 AG072979 (PI David Wolk, MD), P30 AG072972 (PI Charles DeCarli, MD), P30 AG072976 (PI Andrew Saykin, PsyD), P30 AG072975 (PI David Bennett, MD), P30 AG072978 (PI Neil Kowall, MD), P30 AG072977 (PI Robert Vassar, PhD), P30 AG066519 (PI Frank LaFerla, PhD), P30 AG062677 (PI Ronald Petersen, MD, PhD), P30 AG079280 (PI Eric Reiman, MD), P30 AG062422 (PI Gil Rabinovici, MD), P30 AG066511 (PI Allan Levey, MD, PhD), P30 AG072946 (PI Linda Van Eldik, PhD), P30 AG062715 (PI Sanjay Asthana, MD, FRCP), P30 AG072973 (PI Russell Swerdlow, MD), P30 AG066506 (PI Todd Golde, MD, PhD), P30 AG066508 (PI Stephen Strittmatter, MD, PhD), P30 AG066515 (PI Victor Henderson, MD, MS), P30 AG072947 (PI Suzanne Craft, PhD), P30 AG072931 (PI Henry Paulson, MD, PhD), P30 AG066546 (PI Sudha Seshadri, MD), P20 AG068024 (PI Erik Roberson, MD, PhD), P20 AG068053 (PI Justin Miller, PhD), P20 AG068077 (PI Gary Rosenberg, MD), P20 AG068082 (PI Angela Jefferson, PhD), P30 AG072958 (PI Heather Whitson, MD), P30 AG072959 (PI James Leverenz, MD).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ Although people in our sample, at an average of 75 years, cannot be expected to be working (and trading-off the riskiness of a job with the higher wage that accompanies the extra risk) the labour market VSL literature valuation can still be used to provide an approximate value for a QALY. This is because the VSL value that we are using, based on Aldy and Viscusi (2008), is from their oldest cohort, and therefore reflects the preferences from the last time that the people in our sample were working. Note this VSL of 5.09 million USD was for someone aged 62 who had a life expectancy of 23 years. Someone aged 75 has a life expectancy that is 14 years, which is less than 23. Thus, to allow for this life expectancy difference between the time when someone was working and when they were in our sample, the VSL was scaled down to become 3.27 million USD, and this is the valuation that is used to value the QALYs – see also footnote 5.

² It is more usual for data in a panel to use it as the identifying subscripts. But, in our data set, there was a strong overlap between the visit number j and the year t, which enables us to use ij as the unique identifying subscript (for many clients, more than one visit took place in any given year).

³ In a random effects framework, the estimated coefficient α₁ will be unbiased only if the error terms α_i and α_v are assumed to be uncorrelated with all the independent variables and controls in Equation (3).

⁴ The medications are: Ibuprofen, Naproxen, Fenoprofen, Ketoprofen, Sulindac, Indomethacin, Tolmetin, Flurbiprofen, Ketorolac, Meclofenamate, Mefenamic, Nabumetone, Piroxicam, Diclofenac, Etodolac, Oxaprozin, Bromfenac, Diclofenac-Misoprostol, Meloxicam, Lansoprazole-Naproxen, Esomeprazole-Naproxen, Famotidine-Ibuprofen, Aspirin, Diflunisal, Choline Salicylate, Salsalate, Sodium Salicylate, Sodium Thiosalicylate, Magnesium Salicylate, Choline Salicylate-Magnesium Salicylate, ASA/Citric Acid/Na Bicarb, Al Hydroxide/ASA/Ca Carbonate/Mg Hydroxide, Celecoxib, Rofecoxib, Valdecoxib, APAP/ASA/caffeine/salicylamide, APAP/ASA/Caffeine, APAP/Al Hydroxide/ASA/Caffeine/Mg Hydroxide, ASA/Caffeine/Salicylamide, Aspirin-Meprobamate, Aspirin-Caffeine, Aspirin-Phenyltoloxamine, Magnesium Salicylate-Phenyltoloxamine, ASA/Butalbital/Caffeine, Aspirin/Butalbital, Aspirin-DiphenhydrAMINE, DiphenhydrAMINE-Magnesium Salicylate, Acetaminophen-Salicylamide, APAP/Caffeine/Phenyltoloxamine/Salicylamide, APAP/Phenyltoloxamine/Salicylamide.

APAP/Caffeine/Mg Salicylate/Phenyltoloxamine, DiphenhydrAMINE-Ibuprofen.

APAP/Caffeine/Magnesium Salicylate, Acetaminophen-Aspirin, Caffeine-Magnesium Salicylate, APAP/Magnesium Salicylate/Pamabrom. Note that APAP is the abbreviation for Acetaminophen, and Celecoxib (Celebrex) is the only COX-2 inhibitor available in the US.

⁵ This figure was estimated as follows. In the CBA of FDA-approved-medications, the Value of a Statistical Life (VSL) of 5.09 million USD was used taken from Aldy and Viscusi (2008). This value, when discounted for 14 years (which is the life expectancy for those aged 75 years in our sample) at a 3% rate, became 3.27 million USD in 2000 prices (the base year for both benefit and cost estimates). However, for this CBA of the NSAID, it will follow the Greving et al. (2008) timeline going back in time to an age of 55 years for beginning taking the NSAID, and ending up at the NACC sample average age of 75 years. This makes the timeline 20 years. As VSLY = VSL/LY, and with LY = 20, this made the $\overline{\mathit{VSLY}}$ = 163,500 USD.

⁶ With the average inflation rate between 2000 and 2024 being 2.54%, the cumulative price change was 82.45%. Multiplying the 2002 net-benefits by 1.8245 produces the 2024 estimate given in the text.

⁷ Although the evidence shows clearly that the net-benefits for the FDA-Approved medicines for dementia and the beta-blockers are still negative in this paper, even in the presence of someone also taking the NSAID, this is less clear for the antipsychotics. This is because the direct increased mortality effect due to the antipsychotics was only significant at the 6.8% level in Table 3. However, this reservation is not conclusive because in the individual CBA of the Antibiotics in Brent (2024b), an indirect adverse mortality effect was found. This involved the medicine causing dementia to rise, and raising dementia increases mortality. This indirect adverse mortality effect was also found in this paper using all the new controls now used. Dementia symptoms rose by 1.887 points, and a 1 point increase in dementia raises mortality by 0.0119 (both estimates significant at well below the 1% level). Mortality rose indirectly by 0.0224, which is even greater than the 0.0154 direct effect in Table 4. Thus, the mortality estimate for antipsychotics in Table 3 turns out to be a conservative estimate.

⁸ The benefit-cost ratio was 3.31 for the NSAID, while it was 57.12 for vision correction, 11.09 for Medicare eligibility and 3.69 for avoiding nursing homes.

⁹ Although the net-benefits for nursing homes was also negative, and only became large and positive by avoiding living in them, this intervention was different from avoiding taking the pharmaceuticals. This was because positive costs still had to be incurred accommodating residents somewhere else than the nursing homes. This meant there was no cost-savings avoiding nursing homes, as there would be when pharmaceuticals costs would no longer be required, if taking them were now avoided.