Driving forces behind increasing cardiovascular drug utilization: a dynamic pharmacoepidemiological model

Abstract

WHAT IS ALREADY KNOWN ABOUT THIS SUBJECT

• Several studies indicate that switch to more expensive drugs and increasing treatment intensity, rather than population ageing have been responsible for rising drug expenditures during the 1990s.

• Little is known about the driving forces behind the increasing treatment intensity with cardiovascular drugs.

WHAT THIS STUDY ADDS

• This study provides a new pharmacoepidemiological method to analyse drug utilization trends, applying dispensing data at the individual level.

• The suggested semi-Markov model allows for quantification of the influence of changing incidence, discontinuation and user mortality on rising treatment prevalence.

• Increasing treatment incidence was the main driver behind rising treatment prevalence for most cardiovascular drug categories.

• Whereas declining discontinuation drove some of the growth, declining mortality among drug users had little influence.

AIMS

To investigate the driving forces behind increasing utilization of cardiovascular drugs.

METHODS

Using register data, all Danish residents as of 1 January 1996 were followed until 2006. Cohort members were censored at death or emigration. Cardiovascular drug utilization on the individual level was traced, applying registered out-of-hospital dispensing. The impact of population ageing on cardiovascular drug utilization was investigated using standardized intensities and prevalences. Based on a three-state (untreated, treated and dead) semi-Markov model, we explored to what extent increasing treatment prevalence was driven by changing incidence, discontinuation and mortality. Expected treatment prevalences were modelled, applying stratum-specific cohort prevalence in 1996 along with incidence, discontinuation and drug user mortality either throughout 1996–2004 or at fixed 1996 levels.

RESULTS

Treatment prevalence (ages ≥20 years) with cardiovascular drugs increased by 39% during 1996–2005 from 192.4 to 256.9 per 1000 inhabitants (95% confidence interval 256.5, 257.3). Treatment intensity grew by 109% from 272 to 569 defined daily doses 1000⁻¹ day⁻¹. Population ‘middle-ageing’ accounted for 11.5 and 20.3%, respectively. Increasing treatment incidence was the main driver of the rising treatment prevalence in most drug categories. Declining discontinuation drove some of the growth, declining drug user mortality less. Even with fixed incidence in the model, treatment prevalence continued to increase.

CONCLUSIONS

Age-related increases in treatment intensity and prevalence, rather than population ageing, drove the increasing treatment intensity with cardiovascular drugs. Increasing treatment prevalence in subgroups was primarily caused by increasing incidence. Due to pharmacoepidemiological disequilibrium, treatment prevalence will continue to grow even with unchanged incidence.

Introduction

Several studies from Western countries have shown increasing utilization of cardiovascular drugs among nonhospitalized persons, both population-wide and in selected groups (e.g. diseased individuals). In contrast to recent studies with focus on expenditures and quantities, we explored the driving forces behind increasing population-wide drug utilization.

Ageing of the population (i.e. the increasing proportion of elderly and middle-aged) has been pointed out as a major driver of rising healthcare expenditure, but analyses of both drug and other healthcare spending indicate that population ageing may not be the most important factor [1–12]. Studies during the 1990s in Sweden [1] and British Columbia [2, 3] demonstrated that switch to more expensive innovative drugs along with increasing drug quantity per inhabitant have been the main driver behind the expenditure growth, due mainly to changing utilization pattern of drugs affecting the cardiovascular system, central nervous system and gastrointestinal tract.

In line with trends in other Western countries, the aggregate Danish Medical Product Statistics [13, 14] reports an increasing out-of-hospital treatment intensity [defined daily dose (DDD) per 1000 inhabitants day⁻¹] from 851 to 1214 DDD 1000⁻¹ day⁻¹ in the period from 1996 to 2005. For cardiovascular drugs alone, treatment intensity grew from 200 to 397 DDD 1000⁻¹ day⁻¹, accounting for about 55% of the total increase.

Despite the health policy and research relevance of the issue, the underlying driving forces behind the rising treatment intensity with cardiovascular drugs are to a large extent unexplored. There are several potential drivers: population ageing, decreasing mortality among drug users, rising treatment incidence (reflecting increased morbidity and/or changing prescribing behaviour), longer treatment duration (reflecting, for example, changing recommendations and/or better adherence) and increased prescribed daily dose.

The aim of the study was to investigate driving forces behind observed increases in utilization of cardiovascular drugs, both in terms of ageing and changing drug utilization patterns. We developed a dynamic pharmacoepidemiological model for exploring to what extent changes in incidence, discontinuation and drug user mortality can explain observed changes in treatment prevalence.

Methods

Data

From two nationwide registers maintained by Statistics Denmark, we retrieved data on demographics and dispensed cardiovascular prescription drugs. Records on the individual level from the demographic register and the dispensing database were linked by means of the Danish Civil Registration Number, replaced by an anonymous unique person code. The demographic register information contains information on gender, year of birth and, if applicable, date of death or emigration.

Danish inhabitants as of 1 January 1996 identified in the demographic register were followed from this date to 31 December 2005. In order to categorize cohort members as prevalent or nonprevalent drug users at entry into the cohort, we used the calendar year 1995 as run-in period and required persons to be fully observable throughout this period [15–17]. Drug utilization analysis was each calendar year limited to individuals aged ≥20 years. Individuals were censored at death or at the date of first emigration (Table 1). For individuals entering the cohort because they reached the age of 20 years, the previous calendar year was used as run-in year.

Table 1.

The Danish study cohort (1996–2005): members by gender and age group

Number of cohort members present by 1 January								Number censored (1996–2004)
Gender	Age	1996	1998	2000	2002	2004	2005	Deaths	Emigrated
Male	0–19	553 846	485 650	422 835	366 110	311 644	283 431	1 140	11 405
	20–39	776 275	749 992	721 328	685 945	645 748	626 508	6 963	56 327
	40–54	577 369	572 571	558 825	544 461	544 983	547 068	20 474	13 788
	55–64	263 130	282 131	309 280	335 510	349 021	353 592	31 804	3 979
	65–74	199 482	196 533	196 360	200 118	206 825	211 569	56 032	1 299
	75–84	118 632	118 393	119 458	120 302	121 110	120 962	83 089	264
	+85	31 893	33 385	34 215	34 725	35 299	36 494	54 400	39
Male	All	2 520 627	2 438 655	2 362 301	2 287 171	2 214 630	2 179 624	253 902	87 101
Female	0–19	527 694	462 870	402 632	348 232	296 050	268 929	545	15 673
	20–39	740 576	713 657	684 774	650 689	612 575	594 116	3 073	52 614
	40–54	562 137	558 701	547 324	534 760	535 327	537 419	12 740	8 769
	55–64	270 467	287 332	311 660	335 921	350 286	355 542	21 384	2 557
	65–74	233 816	227 725	224 050	225 361	229 632	233 199	42 817	906
	75–84	179 755	178 473	177 918	176 041	173 336	170 674	81 098	340
	+85	76 443	79 443	81 530	82 299	82 079	84 418	103 993	82
Female	All	2 590 888	2 508 201	2 429 888	2 353 303	2 279 285	2 244 297	265 650	80 941
Totals	–	5 111 515	4 946 856	4 792 189	4 640 474	4 493 915	4 423 921	519 552	168 042

The dispensing database contains information on all out-of-hospital purchases of prescription drugs at Danish pharmacies, including those of nursing home residents. The following information was recorded for each redemption: the anonymous person code, date of dispensing, identification of the drug by means of Anatomical Therapeutic Chemical (ATC) classification [18], number of packages, DDD [18] and pharmacy retail price. The records contain information neither on prescribed daily doses nor on the medical indication for prescription.

We included as cardiovascular drugs both those in ATC group C (cardiovascular drugs) and B01 (antithrombotic agents) that are mainly used for the treatment of cardiovascular conditions (Table 2). Drugs were divided into 17 categories according to therapeutic subgroups based on ATC codes. Drugs from ATC group C, not used for chronic treatment of major cardiovascular indications, were excluded: C01C (cardiac stimulants excluding cardiac glycosides), C01E (other cardiac preparations), C04 (peripheral vasodilators) and C05 (vasoprotectives).

Table 2.

Cardiovascular drugs by category^*

Drug category	ATC* classification
1.Vitamin K antagonists	B01AA
2.Heparin	B01AB
3.Platelet aggregation inhibitors (excl. heparin and acetylsalicylic acid)	B01AC (-04,-07,-11)
4.Acetylsalicylic acid	B01AC06
5.Cardiac glycosides (e.g. digitalis)	C01A
6.Antiarrythmics (class I and III)	C01B
7.Vasodilators (used for cardiac diseases, e.g. nitrates)	C01D
8.Antihypertensives (e.g. antiadrenergic agents)	C02
9.Low-ceiling diuretics (thiazides and non-thiazides)	C03A, C03B
10.High-ceiling diuretics	C03C
11.Potassium-sparing agents (alone and in combination with diuretics)	C03D, C03E
12.β-Blocking agents (alone and in combination with other agents)	C07
13.Calcium channel blockers (selective/nonselective; alone/combination with diuretics)	C08
14.ACE inhibitors (plain and combinations)	C09A, C09B*
15.Angiotensin II antagonists (plain and combinations)	C09C, C09D
16.Statins (lipid-modifying agents: HMG CoA reductase-inhibitors.)	C10AA*
17.Other lipid-modifying agents (fibrates, bile acid sequestrants, nicotinic acids)	C10AB, C10AC, C10AD

^*Cardiovascular drugs: ATC group C and B01 (antithrombotic agents). Drugs from ATC group C not used for chronic treatment of major cardiovascular indications were excluded: C01C (cardiac stimulants excl. cardiac glycosides), C01E (other cardiac preparations), C04 (peripheral vasodilators) and C05 (vasoprotectives). ATC, Anatomical Therapeutic Chemical code; ACE, angiotensin-converting enzyme; HMG CoA, 3-hydroxy-3-methylglutaryl coenzyme A.

The influence of population ageing

Because our cohort design allowed for emigration but not for immigration and thus by itself would lead to gradual changes in age distribution, the population-wide treatment intensity was age and gender standardized to the total Danish population aged ≥20 years. We used direct standardization [19], applying yearly age- and gender-specific cohort treatment intensities together with age and gender weights of the Danish population (1996–2005).

The impact of population ageing on cardiovascular treatment intensity was analysed by comparing the yearly standardized treatment intensity with the intensity standardized to the 1996 population. In the following we refer to the yearly standardized intensities as population intensities, whereas the 1996-standardized intensities are referred to as standardized intensities. Similarly, we use the concepts population prevalences and standardized prevalences. Thus, the influence of ageing on treatment intensity was calculated as (population intensity 2005 – standardized intensity 2005)/(population intensity 2005 – population intensity 1996). The influence of ageing on treatment prevalence was analysed analogously. As population ageing corresponds to the combined effect of an increasing remaining life time (decreasing mortality) and the ageing of the large post-war baby-boom cohorts, it is not possible from the analyses described above to infer to what extent each of the two components contributes to rising drug utilization. Below, we explore the impact of decreasing drug user mortality on changing treatment prevalences.

The pharmacoepidemiological model

We used a three-state (untreated, treated and dead) semi-Markov model [20] to analyse the dynamics of drug utilization and to explore to what extent increasing treatment prevalence of particular cardiovascular drugs was driven by changing incidence, discontinuation and mortality proportions. Potential transitions between two calendar years were from untreated to treated, the reverse, and from either untreated or treated to death, see Figure 1. The model is based on the fact that the number of individuals in the prevalent treatment state (users) at the end of the year equals the observed number at the beginning plus the number of surviving incident users occurring during the year, minus the number of users who discontinued treatment or died during the year [17]. Ignoring migration, the relationship is exact.

Figure 1.

Paths and transitions in the dynamic pharmacoepidemiological model

Individuals, who by the index date (1 January, each calendar year) had at least one prescription of a particular drug during the preceding 365 days, were assigned to the treated state (prevalent treatment state). Individuals in the treated state (drug users) with no observed dispensing of the drug between the index date and the following 365 days were assigned to the untreated state at the following calendar year provided they were still alive. This transition from treated to untreated state was considered an event of treatment discontinuation. Non-users with observed dispensing of the particular drug during the 365 days following the index date were assigned to the treated state at the following year, if still alive. This transition from the untreated to the treated state was considered an event of incident treatment (see Figure 1).

Stratified by gender and 5-year age groups, the following calendar yearly measures were calculated: prevalence proportion (number of users at the beginning of the year/total number at the beginning of the year), incidence proportion (number of incident users during the year/number of non-users at the beginning of the year), discontinuation proportion (number of discontinued users during the year/number of users at the beginning of the year) and mortality proportions stratified by treatment state [number of user(non-user) deaths during the year/number of users(non-users) at the beginning of the year].

Note that the applied cumulative ‘net’ incidence proportion deviates slightly from the definition of a cumulative incidence proportion (‘Number of subjects developing disease during a time interval/number of subjects who entered the time interval’, see [21]), in that, as numerator, it uses the number of incident drug users surviving to the end of the time interval rather than the total number of incident drug users during the time interval.

Using the model, age- and gender-specific treatment prevalences in the Danish population at the end of each calendar year were calculated, applying the stratum-specific cohort prevalence proportions as of 1996 along with the calculated stratum-specific cohort incidence, discontinuation and drug user mortality proportions throughout 1996–2004. We accounted for migration by making the calculated stratum-specific model prevalences at the beginning of a calendar year correspond to the prevalences at the end of the preceding year, assuming pragmatically that immigrants and emigrants are similar with respect to treatment status and that net migrants have the same stratum-specific treatment prevalence as the nonmigrating population.

The modelled population-wide treatment prevalence, designated model prevalence, was calculated, applying the above described stratum-specific model prevalences along with weights of the Danish population (aged ≥30 years).

We investigated to what extent changes in incidence, discontinuation and user mortality could explain observed changes in population-wide treatment prevalence of a particular drug category, by fixing one measure at a time at the level of 1996 (scenario prevalences). In a few strata with <10 users (notably in the heparin group), we replaced the 1996 point estimate of discontinuation and user mortality proportion with the mean proportions over the first four observation years. The influence of changing pharmacoepidemiological parameters on changing treatment prevalence during 1996–2005 was calculated as (model prevalence 2005 – scenario prevalence 2005)/(model prevalence 2005 – model prevalence 1996).

All data analyses were performed using Stata Release 9 (StataCorp LP, College Station, TX, USA).

Sensitivity analysis

Because the cohort experienced emigration, but not immigration, there was a gradual dilution of the younger segments of the cohort (see Table 1). To evaluate the extent of possible bias, we performed a simple sensitivity analysis in which 90% of the emigrated cohort members aged <40 years re-entered the drug utilization analysis as non-drug users when otherwise censored (i.e. we effectively adjusted the numerator towards its likely true value). Sensitivity analysis revealed that although treatment intensity and prevalence among persons aged <40 years are likely to be overestimated (9% in 2005), population-wide estimates are largely unaffected by the dilution of the younger segment (<0.2% in 2005).

Analytical methods

For each drug category a 95% confidence interval (CI) of the 2005 population treatment prevalence was calculated, applying the following formula [22]:

where w_i=N_i/N = weight of stratum i (N_i) in the Danish population (N) as of 2005, p_i = n_i/N_i = treatment prevalence in stratum i with n_i drug users, and p = Σw_i × p_i = population treatment prevalence.

We evaluated model accuracy and precision for each drug category. The accuracy of the model was evaluated by comparing model prevalence with population prevalence in 2005 (age ≥30 years). Differences were in the interval −0.2 to +0.1 per 1000 except for total cardiovascular drug prevalence where the deviation was +/−0.5 per 1000.

The precision of the model was evaluated by running it 1000 times, substituting the yearly proportions (prevalence in 1996, yearly incidence, discontinuation and mortality) by random draws from a binominal distribution with the corresponding parameters. The 95% model prediction interval was calculated as the 2.5 and 97.5 percentile of the distributions emerging from the 1000 runs. These prediction intervals were almost identical to the CIs of the 2005 population prevalence calculated with the method described above [22], although larger as to stratum-specific predictions.

Ethical considerations

Admission to data was provided and secured through collaboration between the University of Southern Denmark and Statistics Denmark. The latter approved the study. No person identifiers were provided to the researchers.

Results

The influence of ageing

Between 1996 and 2005, population treatment intensity with all cardiovascular drugs increased from 272 to 570 DDD 1000⁻¹ day⁻¹, whereas standardized cardiovascular treatment intensity increased from 272 to 535 DDD 1000⁻¹ day⁻¹ (Table 3), corresponding to an increase of 109 and 97%, respectively. Thus, ageing of the population accounted for 12% of the observed growth in cardiovascular treatment intensity. During the same period population treatment prevalence with cardiovascular drugs increased by 40% from 157 to 219 1000⁻¹, whereas the standardized prevalence increased by 31% from 157 to 206 1000⁻¹. Population ageing accounted for 20% of the growing treatment prevalence.

Table 3.

Treatment intensity and prevalence: population and standardized drug utilization measures (ages ≥20 years)

	Treatment intensity (DDD 1000−1 day−1)				Treatment prevalence (users 1000−1)
	1996		2005		1996		2005
Drug category		Population	Standardized*	Ageing (%)†		Population	Standardized*	Ageing (%)†	Growth group‡
1.Vitamin K antagonists	3.6	7.9	7.5	9	5.9	13.5	13.0	7	A
2.Heparin	0.0	0.2	0.2	–	0.1	0.4	0.3	–	C
3.Platelet aggregation inhibitors	0.1	10.3	9.7	6	0.1	12.9	12.1	6	A
4.Acetylsalicylic acid	22.8	52.9	49.9	10	24.9	56.5	53.4	10	A
5.Cardiac glycosides	10.7	7.1	6.8	–	16.5	11.4	10.9	–	B
6.Antiarrythmics	1.4	1.8	1.7	–	1.7	2.1	2.0	–	C
7.Vasodilators	13.0	11.9	11.3	–	19.7	17.2	16.4	–	B
8.Antihypertensives	1.3	3.6	3.4	–	2.8	4.2	4.0	–	C
9.Low-ceiling diuretics	47.0	62.7	59.3	21	47.4	71.3	67.5	16	A
10.High-ceiling diuretics	61.2	62.6	59.5	–	36.9	36.5	34.7	–	B
11.Potassium-sparing agents	15.4	11.4	10.7	–	19.9	17.5	16.6	–	B
12.β-Blocking agents	20.2	39.9	37.5	12	38.1	70.0	66.3	12	A
13.Calcium channel blockers	37.6	63.1	59.3	15	38.6	51.8	48.9	22	A
14.ACE inhibitors	29.0	77.4	72.4	10	26.0	54.4	51.0	12	A
15.Angiotensin II antagonists	3.5	43.7	40.9	7	3.2	38.1	35.7	7	A
16.Statins	4.9	112.6	104.6	7	4.4	57.1	53.3	7	A
17.Other lipid-modifying agents	0.8	0.7	0.6	–	1.5	1.4	1.3	–	C
All cardiovascular drugs	272.4	569.6	535.3	12	157.3	219.0	206.5	20

^*Applying as weights the gender and age distribution of the Danish population in 1996.

^†The impact of ageing = (Population measure 2005 – standardized measure 2005)/(population measure 2005 – population measure 1996).

^‡A, considerable growth; B, decrease; or C, inconsiderable change in treatment prevalence. DDD, defined daily dose; ACE, angiotensin-converting enzyme.

During 1996–2005, the fraction of middle-aged (50–64) expanded by 18% at the expense of a declining fraction of younger adults (20–49) (see Table 4). The elderly (65+) as fraction of the population remained almost constant. Yet, the stratum-specific treatment prevalence with the group of cardiovascular drugs rose for all ages and with about 100 per 1000 at ages >50 years. On the other hand, treatment intensity in absolute terms rose substantially more among persons aged ≥65 years than among the middle-aged.

Table 4.

Demographic changes and age-related utilization of cardiovascular drugs

	Fractions of total population* (%)		Prevalence (user 1000−1)			Intensity (DDD 1000−1 day−1)
Age	1996	2005	1996	2005 (increase, %)		1996	2005 (increase, %)
20–49	58.1	54.0	40	51	(27)	44	86	(94)
50–64	22.1	26.1	199	283	(42)	329	723	(120)
65–84	17.6	17.4	431	569	(32)	839	1658	(98)
85+	2.2	2.4	638	752	(18)	1187	1859	(57)

^*Fraction of the Danish Population aged ≥20 years. DDD, defined daily dose.

Treatment intensities vs. prevalences

Between 1996 and 2005, treatment intensity with angiotensin-converting enzyme (ACE) inhibitors, angiotensin II antagonists and statins increased by 48, 40 and 108 DDD 1000⁻¹ day⁻¹, respectively (Table 3). The rising treatment intensity with these three ‘major growth’ drug categories represented two-thirds of the total increase of 298 DDD 1000⁻¹ day⁻¹ in cardiovascular treatment intensity.

Treatment prevalence of all cardiovascular drug categories followed the up- or downward trends of treatment intensity, but with different patterns. For four drug categories with substantial increases in standardized treatment intensity (platelet aggregation inhibitors, calcium channel blockers, ACE inhibitors and statins), treatment intensity rose considerably more than treatment prevalence. For example, standardized treatment intensity with statins increased by a factor of 20, whereas treatment prevalence increased by a factor of 11. Other drug categories (vitamin K antagonists and low-ceiling diuretics) had more pronounced increases in treatment prevalence than in treatment intensity. For example, treatment intensity with vitamin K antagonists increased by 108%, whereas the prevalence rose by 121%.

Drug utilization trends

The 17 drug categories can be divided roughly into three groups, according to utilization trends (see Table 3, last column):

1. Nine drug categories with considerable growth in standardized treatment prevalence, i.e. a growth of >25% along with treatment prevalence in 2005 >5 per 1000 inhabitants. A general trend of increasing treatment incidence was observed for all nine drug categories.

2. Four drug categories with decreasing standardized prevalence, mainly explained by decreasing incidence.

3. Four drug categories with almost unchanged standardized treatment prevalence or inconsiderable growth.

Pharmacoepidemiological model analysis

In Table 5 population prevalences, model prevalences and scenario prevalences are shown for all cardiovascular drug categories including the entire group of cardiovascular drugs. The driving pharmacoepidemiological forces behind changing drug treatment prevalences can be inferred by comparing scenario prevalences with the corresponding model prevalences.

Table 5.

Population treatment prevalences (age ≥30 years), model prevalences and scenario prevalences fixing incidence, discontinuation or mortality proportion

		Population prevalence†			Model and scenario prevalences for 2005
					Model	Scenario prevalences (influence (%))§
Drug category (growth group*)		1996	2005 (95% CI)		prevalences*	Incidence		Discontinuation		Mortality
1.Vitamin K antagonists	(A)	7.1	15.9	(15.7–16.0)	15.8	11.8	(46.2)	13.4	(27.8)	15.2	(7.2)
2.Heparin group	(C)	0.1	0.4	(0.4–0.4)	0.4	0.1	(91.4)	0.4	(−1.5)	0.4	(2.6)
3.Platelet aggregation inhibitors	(A)	0.2	15.2	(15.1–15.4)	15.2	0.6	(97.4)	14.0	(8.1)	15.0	(1.4)
4.Acetylsalicylic acid	(A)	30.9	66.9	(66.6–67.1)	66.8	48.4	(51.2)	58.8	(22.4)	65.2	(4.5)
5.Cardiac glycosides	(B)	20.4	13.5	(13.4–13.7)	13.5	18.4	(71.3)	14.5	(14.4)	13.0	(−7.5)
6.Antiarrythmics	(C)	2.1	2.5	(2.4–2.5)	2.5	1.9	(155.3)	2.7	(−63.7)	2.4	(19.5)
7.Vasodilators	(B)	24.4	20.4	(20.3–20.6)	20.4	24.4	(100.1)	22.4	(51.2)	19.8	(−14.4)
8.Antihypertensives	(C)	3.4	5.0	(4.9–5.0)	5.0	3.0	(121.1)	5.7	(−48.9)	4.8	(9.9)
9.Low-ceiling diuretics	(A)	58.5	84.2	(83.9–84.5)	84.0	66.8	(67.6)	78.2	(22.7)	82.6	(5.6)
10.High-ceiling diuretics	(B)	45.5	43.1	(42.9–43.3)	43.0	47.2	(166.1)	41.4	(−64.9)	42.2	(−32.7)
11.Potassium-sparing agents	(B)	24.5	20.6	(20.5–20.8)	20.6	21.1	(12.2)	19.1	(−37.4)	20.8	(6.0)
12.β-Blocking agents	(A)	45.7	81.4	(81.1–81.7)	81.2	61.0	(56.9)	73.4	(22.1)	81.0	(0.6)
13.Calcium channel blockers	(A)	47.6	61.2	(61.0–61.4)	61.1	60.7	(3.2)	57.5	(27.0)	59.6	(11.1)
14.ACE inhibitors	(A)	32.0	64.2	(63.9–64.4)	64.1	46.6	(54.5)	60.9	(10.1)	62.4	(5.5)
15.Angiotensin II antagonists	(A)	3.9	45.0	(44.8–45.3)	45.0	19.2	(62.7)	31.5	(32.8)	44.7	(0.7)
16.Statins	(A)	5.4	67.6	(67.3–67.8)	67.5	22.6	(72.3)	68.1	(−1.0)	67.4	(0.2)
17.Other lipid-modifying agents	(C)	1.8	1.6	(1.5–1.6)	1.6	1.5	(−24.1)	1.8	(94.6)	1.6	(2.2)
All cardiovascular drugs		192.4	256.9	(256.5–257.3)	256.4	234.7	(33.9)	233.7	(35.4)	253.4	(4.7)

^*A, considerable growth, B, decrease, or C, inconsiderable change in treatment prevalence.

^†Population prevalences (users per 1000 inhabitants), applying point estimates of stratum-specific cohort prevalences and weights corresponding to the Danish population (aged ≥30 years, 1996–2005).

^‡Model prevalences (users per 1000 inhabitants): model run for the Danish population (aged ≥30 years, 1996–2005), applying point estimates of stratum-specific cohort incidence, discontinuation and mortality proportion; 95% model prediction confidence interval.

^§Scenario prevalence (users per 1000 inhabitants): model run for the Danish population, fixing stratum-specific incidence, discontinuation or mortality at the levels in 1996. The influence of changing parameters (model prevalence 2005 – scenario prevalence 2005)/(model prevalence 2005 – model prevalence 1996). ACE, angiotensin-converting enzyme.

Figure 2 depicts model prevalence vs. scenario prevalences for the drug categories of angiotensin II antagonists, statins and ACE inhibitors. Figure 2 reveals a substantial influence of both changing treatment incidence and discontinuation on treatment prevalence with angiotensin II antagonists. Had either treatment incidence or discontinuation remained unchanged since 1996, the rise in treatment prevalence with angiotensin II antagonists would have been more modest. Unchanged user mortality since 1996, however, would hardly have changed the treatment prevalence. Therefore increasing treatment incidence and declining treatment discontinuation (the former more than the latter) both acted as driving forces behind the rising treatment prevalence with angiotensin II antagonists, contributing with 63 and 33%, respectively, of the increase, whereas changing user mortality hardly did so (see Table 5).

Figure 2.

Model prevalences versus scenario prevalences, fixing incidence, discontinuation or drug user mortality. Model prevalence (); Scenario, fixed incidence (); Scenario, fixed discontinuation (); Scenario, fixed mortallty ()

Holding statin treatment discontinuation constant as well as user mortality at their levels of 1996 hardly affected the growth in statin treatment prevalence. However, fixing treatment incidence at the level in 1996 brought about a more modest increase in treatment prevalence. Increasing treatment incidence thus accounted for 72% of the rising treatment prevalence with statins, whereas changing discontinuation or user mortality had hardly any impact on this increase.

In contrast, as regards ACE inhibitors, declining mortality, declining discontinuation and increasing incidence contributed to the rising treatment prevalence with 6, 10 and 47%, respectively. Thus, a common characteristic of these three ‘major growth’ drug categories was that increasing treatment incidence proved to be the main pharmacoepidemiological force behind increasing treatment prevalence.

Fixing treatment incidence for high ceiling diuretics (growth group B) at the higher level of 1996 resulted in a scenario prevalence of 47.2 in 2005 compared with the decreasing model prevalence from 45.5 to 43.0. Thus, the decreasing treatment prevalence can be attributed to declining treatment incidence. Declining treatment discontinuation and user mortality did, however, to some extent counteract the impact of the declining treatment incidence.

As regards the entire group of cardiovascular drugs, increasing treatment incidence and declining treatment discontinuation contributed with 34 and 35%, respectively, of the increasing treatment prevalence. Declining user mortality drove 5% of the increase. The influence of declining user mortality was most pronounced among men and was increasing with age from the age of 55 (men 4.5–13.3%, women 1.7–7.4%, data not shown).

Discussion

In contrast to recent analyses of population-wide drug utilization, we focused on both pharmacoepidemiological and demographic driving forces behind the increasing utilization of cardiovascular drugs rather than on drug expenditures and quantities. We thus examined some of the unexplored issues emerging from recent published studies on drug utilization [1, 2, 23].

Population ageing

In line with studies from Sweden and British Columbia [2, 3, 23, 24], we found that increases in stratum-specific treatment intensities (and prevalences) rather than population ageing were the main driving forces behind the rising utilization of cardiovascular drugs. Our finding that population ageing, i.e. the ‘middle-ageing’ of the post-war baby-boom generation, accounted for substantially more of the increasing treatment prevalence than of treatment intensity (20% vs. 12%) was explained by the fact that treatment intensity in absolute terms grew substantially more among elderly (age ≥65 years) than among middle-aged, while the growth in treatment prevalence was about the same.

Treatment intensities vs. prevalences

In line with a study from British Columbia [2], we found that both increasing age-specific treatment prevalences and treatment intensities contributed to the increasing treatment intensity with cardiovascular drugs.

The finding, that cardiovascular treatment intensity grew faster than treatment prevalence, combined with a growing treatment prevalence of many drug categories, points towards an increasing tendency of polytherapy within the group of cardiovascular drugs (i.e. an increasing number of different drug categories per drug user), rather than an increasing prescribed/dispensed daily dose of particular drug categories.

However, a trend of increasing prescribed daily dose with statins was found in a recent study of statins utilization (with access to information on prescribed daily doses) in EU member states and Norway during 1997–2003 [25]. The authors link this trend to three potential circumstances: statins are recommended for still lower levels of hypercholesterolaemia, accordingly treatment goals have been sharpened, and finally, that DDD measures are not equipotent, i.e. the lipid-lowering effect per unit varies for the different statins. Without information on prescribed daily doses in the present study, we focused on the fact that treatment intensity with four drug categories (statins, ACE inhibitors, calcium channels blockers and platelet aggregation inhibitors) grew faster than treatment prevalence, indicating increases in average prescribed daily dose of these drugs.

Pharmacoepidemiological model analyses

The dynamic model uncovered the main pharmacoepidemiological drivers of increasing treatment prevalence, regarding both the individual cardiovascular drug categories and the entire group of cardiovascular drugs.

Categories of cardiovascular drugs

The pattern of driving forces proved dependent on cardiovascular drug category. However, increasing treatment incidence was the most important driving force in seven out of nine drug categories with increasing treatment prevalence. Correspondingly, declining treatment incidence was the most important driver for three out of four categories with declining treatment prevalence.

Following this pattern, rising treatment incidence was the most important driving force for the three drug categories with major growth – statins, angiotensin II antagonists and ACE inhibitors. Increasing treatment incidence was almost exclusively responsible for the increasing treatment prevalence of statins.

The model analyses shed light on the fact that rising treatment prevalence does not necessarily follow from rising treatment incidence: if pharmacoepidemiological disequilibrium exists (i.e. the number of people initiating drug therapy exceeds the number ceasing therapy), treatment prevalence will increase even without rising treatment incidence, as exemplified in the model analysis of statins.

Despite concern about noncompliance with lifelong statin therapy [26–30], changing treatment discontinuation with statins had virtually no impact on observed treatment prevalence. However, our measure of treatment discontinuation (a 1-year cumulative proportion) only crudely reflects treatment duration. The analysis would be likely to benefit from a detailed definition of discontinuation and paused treatment [15, 16, 31–33], including knowledge or assumptions about the drug quantity redeemed and the prescribed daily dose [34]. Although there are different theoretical approaches, survival analyses are normally applied to study compliance and treatment persistence [35–38].

The group of cardiovascular drugs

Both incidence and case fatality of ischaemic heart disease have been declining in Denmark during recent decades [39]. Yet, according to the present analysis increasing treatment incidence and decreasing treatment discontinuation accounted for 34 and 35%, respectively, of the growth in treatment prevalence with respect to the entire group of cardiovascular drugs. If the incidence of cardiovascular morbidity (i.e. not merely ischaemic heart disease) was indeed constant or falling during the period, it may be concluded that the growing cardiovascular drug treatment prevalence is mainly driven by forces internal to the healthcare system (medico-political decisions) and/or changing public demands on the healthcare system.

Declining mortality among users of cardiovascular drugs accounted for 5% of the increasing overall treatment prevalence with the group of cardiovascular drugs. However, declining drug user mortality (most pronounced among men) might both be attributable to improved cardiovascular care (including improved drug treatment) and to the ‘healthy drug user effect’, i.e. prescribing cardiovascular drugs to still healthier persons combined with a general trend of declining mortality.

The drugs exhibiting a considerable increase in treatment intensity and prevalence in this analysis are responsible for a large proportion of drug costs. During the observation period expenditures (pharmacy retail prices) on cardiovascular drugs (ATC group C) jumped from 161 to €238 million, an increase of 48%. Even with a pronounced decrease in expenditure per DDD (as regards statins and ACE inhibitors), spending on the three drug categories with major growth accounted for 32 and 55% of the total cardiovascular expenditure in 1996 and 2005, respectively [13, 14].

Strength and limitations

Even with a relatively simple pharmacoepidemiological model based on treatment incidence, treatment discontinuation and drug user mortality proportions, overall treatment prevalence was predicted with good accuracy. By means of the model, we have revealed important dynamics behind observed trends in treatments prevalences with respect to cardiovascular drugs for chronic use.

However, selection bias might have arisen, as in this model we could not properly assess treatment status among migrating individuals [40]. The influence on model results is considered negligible, as both new immigrants and emigrants each year constituted only about 0.1% of the population aged >55 years and 1.3% of those aged <55 years. Potentially more serious selection bias could arise due to dilution of the younger segment in the cohort (the accumulated impact of emigration). However, sensitivity analysis revealed that even though the dilution led to overestimation of drug utilization among younger persons, it had virtually zero influence on estimated population-wide measures of drug utilization. This is a consequence of cardiovascular drug utilization among the older being many times greater than among the younger.

Keeping the model ‘nice and simple’ has some limitations. The present model is not likely to be applicable for more detailed analyses of drug utilization patterns such as analyses of discontinuation patterns or treatment duration patterns. Moreover, neither the pharmacoepidemiological model nor the methods used to study the impact of ageing apply directly to analysing polytherapy with cardiovascular drugs.

Future research

From a methodological point of view, the model could be improved by applying rates instead of proportions and by including all potential consecutive events on the path from user to nonuser or vice versa. From a more practical point of view, the application of the model could be enhanced by encompassing polytherapy. Register-based data on cardiovascular morbidity could be included in the analysis, shedding light on the impact of changing indications on the cardiovascular treatment prevalence. We plan to improve our model to cover these aspects.

Conclusion

During 1996–2005, treatment intensity and prevalence with cardiovascular drugs in Denmark rose by 109 and 40%, respectively. Three major drug categories, ACE inhibitors, angiotensin II antagonists and statins, accounted for two-thirds of the growth. In line with recent drug utilization studies, we found that stratum-specific increases in treatment intensity and prevalence rather than population ageing drove the overall increase in treatment intensity with cardiovascular drugs.

Increasing treatment incidence was the main pharmacoepidemiological force behind rising treatment prevalence for most cardiovascular drug categories. While declining discontinuation drove some of the growth, declining drug user mortality contributed to a lesser extent. Due to pharmacoepidemiological disequilibrium, treatment prevalence of many cardiovascular drug categories would have continued to increase, even without increases in treatment incidence.

Competing interests

M.A. has participated in research projects funded by AstraZeneca, Lundbeck, Novartis and Nycomed and has received fees for teaching from the Danish Association of the Pharmaceutical Industry.

The study was funded by the Danish Research Foundation for General Practice (grant no. 585-04/3813).