METHODS

Selection of patients

All patients first diagnosed with OSAS between 1997 (at least one year after the pharmacy database was established (see below)) and 1999 were included in the study. The diagnosis was confirmed in all patients by polysomnography which included the monitoring of neurophysiological measures (EEG, chin EMG, EOG, anterior tibialis EMG) and cardiorespiratory variables (chest wall motion, abdominal motion, oronasal Pco₂, Spo₂ (by fibreoptic pulse ear oximeter), and ECG). The key measures obtained from polysomnographic monitoring were apnoea/hypopnoea index (AHI) as one indicator of apnoea severity and time spent below Sao₂ 90% (time <90%) as an index of exposure to hypoxaemia.

Patients who, in addition to sleep apnoea, had another sleep disorder such as co-existing narcolepsy and periodic leg movements which did not resolve on treatment of the apnoea were excluded from the study. Patients whose entire health care services were not paid by the province of Manitoba—for example, those not permanently resident in Manitoba, status Indians, and military personal (whose health care is covered by the Canadian Federal Government)—were also excluded. There remained 551 eligible cases for matching with controls and analysis.

Selection of controls

Each OSAS case was matched with a control subject from the general population for sex, area of residence, postal code, age, and physician. The cases were matched to controls using the Manitoba health population registry and the database which maintains records of physician claims for services and hospital admissions on all residents of Manitoba. As most of the cases had a number of physician contacts before the date they were evaluated in the sleep laboratory, the most frequently seen physician in an ambulatory setting in the previous 2 years was taken as the target for the match. All those who saw the same physician in the same time period as each case were identified. One person was chosen at random for each case if several were available. Of the 551 eligible to be matched with a control, 549 matched at least one person for sex, age within 5 years, and physician (contact with case's most frequent physician). The other two cases had no suitable matches. A 5 year age tolerance was allowed in case a 1 year age match could not be obtained; however, the age difference was chosen so that the control was always older than the case. Of the 549 matches, 546 (99.5%) were exact for age.

This research was designed from the onset to protect confidentiality of cases and controls. Only encrypted forms of subjects' personal health information numbers were used. The study was approved by the University of Manitoba human ethics committee and the access and confidentiality committee of Manitoba Health.

Description of the database

All outpatient prescriptions completed in Manitoba are tracked by the Drug Programs Information Network (DPIN) which was established in 1996.¹⁶ All the dispensing pharmacies in the province of Manitoba are linked in real time with the Manitoba health computer system. Each dispensed prescription generates a record that includes a personal identifier (an encrypted version of the person's health information number), the date the prescription was dispensed, the drug identification number (DIN) of the medication, the dosage prescribed, and the number of doses prescribed. In 1999 this database included 707 specific medications (both generic and brand name) for a total of 3328 products. The personal identifier is used for linking to another database¹⁷ which can then yield region of residence, sex, prescribing physician, and physician claims.

The medications were classified according to the World Health Organization Collaborating Centre for Drug Statistics methodology.¹⁸ This classification divides medications into groups based on organ system or therapeutic class. The classification yields anatomical therapeutic chemical classes (ATC). For example, ATC=C is the group of drugs dispensed for cardiovascular disease. This database does not include prescriptions for hospital inpatients, military personnel, or status Indians. The latter two groups are covered by a different funding source. The database has been validated¹⁶ and 92% of the prescriptions in the database matched the original written prescription. It does not include over the counter medications or products purchased in health food stores.

Primary outcome measures

Information was obtained from the database about all prescriptions dispensed to the cases in the 12 months before the diagnosis of OSAS was first confirmed. For each matched control we obtained information about all their prescriptions for the same time frame as their matched case.

Number of prescriptions

The number of prescriptions and the medication prescribed, classified according to anatomical and therapeutic chemical (ATC) criteria according to WHO methodology,¹⁸ was determined.

Defined daily doses (DDD)

The defined daily dose is the mean dose per day for a medication when used in treating its major indication. The DDD was calculated both for the entire population of cases and controls and for the subgroups of case users and control users (who actually presented the prescriptions). It has been used in quantifying intensity of drug use. Defined daily doses are not available for combination drugs.

Days of drug supplied (DDS)

This is the number of days for which medication is supplied. This value differs from DDD in that, for some indications, different doses of a drug might be prescribed. This measure includes combination drugs for which DDDs are not available.

Cost

The cost in Canadian dollars ($1Can=$0.68 US) of medications (per prescription and per year) for the population of cases and controls was calculated.

Predictors of medication use

The following independent variables were evaluated to see whether they significantly impacted on medication use: age, sex, body mass index (BMI), apnoea/hypopnoea index (AHI), percentage time below an Sao₂ of 90% during sleep (time <90%), and Epworth sleepiness scale (ESS), a subjective measure of sleepiness.¹⁹

Statistical methods

Odds ratios comparing drug utilisation among cases and controls were obtained from the Cochran-Mantel-Haenzel test allowing for matching. Paired t tests were used to compare cases and controls on continuous variables (days supplied, total cost). Among the cases, logistic regression using stepwise elimination was used to predict drug utilisation as a function of AHI, time below Sao₂ 90%, and ESS controlling for age, sex, and BMI. Linear regression was used in a similar way to model the continuous variables. Comparisons between men and women were carried out using unpaired t tests. Probability values of <0.05 were considered significant. Confidence intervals were calculated using standard methods and computer programs.^20,21 Analysis was conducted on a Sun Microsystems work station running SAS version 8.1 (SAS Institute, Cary, North Carolina), and on a PC running Confidence Interval Analysis version 2.03.²¹ When calculating confidence intervals on proportions, the method of Wilson²² was used as recommended.^20,21

RESULTS

A total of 549 cases (401 men) with OSAS and matched controls were eligible for analysis. Their demographic characteristics are shown in table 1. The patients were, on average, middle aged; the men were significantly younger with a significantly lower BMI but higher AHI than the women.

Use of all drugs

From the drug database we analysed all prescriptions dispensed to patients and their controls. At least one prescribed medication was used by 10.2% (95% CI 5.4 to 15.0; p<0.0001) more cases (83.4%) than controls (73.2%) in the 1 year under consideration. Male cases (80%) were more likely to have been prescribed at least one medication than controls (68.3%), the difference being 11.7% (95% CI 5.7 to 17.7; p=0.0001). However, the difference between female cases (92.6%) and controls (86.5%) of being prescribed at least one medication was 6.1% (95% CI –1.0 to 13.3; p=0.09) and did not reach significance.

Different measures of intensity of use per person per year were calculated. Table 2 shows that cases were on more medications, had more prescriptions per year, and had higher DDD and DDS. Furthermore, the total cost of medications was higher in cases than in controls. However, there was no difference between cases and controls in the cost per prescription. The differences between cases and controls were in the same directions for both men and women; for number of drugs (p=0.02) and DDS (p=0.03) the differences were significant, but for the other measures of intensity there was no significant difference in magnitude.

Female cases (92.6%) were more likely to be using at least some medication than male cases (80.1%); the difference (12.5%, 95% CI 6.0 to 17.8; p<0.0001) was statistically significant. Table 2 shows that female cases had 367.4 more DDS than male cases (713.1 v 345.7, p<0.0001). The total cost of medications for female cases was $310.45 more than for males ($634.7 v $324.3, p<0.0001). The cost of medications for female controls was also higher than for male controls.

Use of cardiovascular system drugs

From all drugs we focused on the medications normally used to treat cardiovascular disease (see table 3 and box 1). At least one cardiovascular medication prescription was completed by 16.9% (95% CI 12.1 to 21.7; p<0.0001) more cases (36.6%) than controls (19.7%) in the 1 year under review. Male cases (34.7%) were more likely to have been prescribed at least one cardiovascular medication than controls (18.2%), the difference being 16.5% (95% CI 10.9 to 21.9; p<0.0001). Female cases (41.9%) were more likely to have been prescribed at least one cardiovascular medication than controls (23.7%), with a difference of 18.2% (95% CI 8.4 to 27.6; p<0.0001).

Different measures of intensity of cardiovascular drug use per person per year were calculated. Table 3 shows that cases were on more medications, had more prescriptions per year, and had higher DDD and DDS. Furthermore, the total cost of medications was higher in cases than in controls. However, there was no difference between cases and controls in the cost per prescription. The differences between cases and controls were in the same directions for both men and women and there was no significant difference in magnitude.

Female cases (41.9%) were not more likely to use cardiovascular medications than male cases (34.7%); the difference (7.2%, 95% CI –1.8 to 16.5; p=0.126) was not significant. There was no significant difference between men and women for drug use, DDS, and total cost of cardiovascular drugs.

Because the ATC=C category includes many drug classes, we divided them into three large groups based on the way they are usually used clinically: drugs that reduce blood pressure, drugs that lower lipids, and drugs whose main effect is on the heart or the coronary circulation (box 1). We first examined drugs that are approved and most widely used to lower systemic arterial blood pressure (antihypertensives) in Canada. Clinicians might use these medications not just for controlling high blood pressure, but also for other indications and benefits from reducing blood pressure such as for the treatment of angina pectoris, congestive heart failure, and the secondary prevention of myocardial infarction in high risk patients. There is no way to determine from the database the specific indication for which a drug is prescribed. Antihypertensive drugs were far more likely to be prescribed for cases than for controls (31.7% v 16.4%, p<0.0001). The odds ratio for OSAS patients using antihypertensive agents compared with controls was 2.71 (95% CI 1.96 to 3.77). The cases had a mean of 61.6 more DDS for antihypertensive drugs (126.9 v 65.3, p<0.0001) and spent an average of $47.88 more on antihypertensive drugs than controls ($95.88 v $48.00, p<0.0001). We then calculated the intensity of drug use for each antihypertensive drug category including anti-adrenergic drugs, diuretics, loop diuretics, beta blockers, calcium channel blockers, and angiotensin converting enzyme (ACE) inhibitors. Table 4 shows that cases had significantly more DDS on diuretics, calcium channel blockers, and ACE inhibitors and spent significantly more on diuretics, beta blockers, calcium channel blockers, and ACE inhibitors than controls.

We also estimated the differences in antihypertensive drug use for men and women. At least one antihypertensive drug was prescribed to 28.7% of the male cases and 14.7% of the male controls (OR 2.65, 95% CI 1.78 to 3.93, p<0.0001). Antihypertensives were prescribed to 39.9% of the female cases and 20.9% of the female controls (OR 2.87, 95% CI 1.59 to 5.16, p=0.0002). We calculated the relative likelihood of female cases versus male cases having a prescription for each antihypertensive drug, other cardiovascular drugs, and cholesterol and triglyceride reducers. Women were more likely to be dispensed prescriptions for diuretics and ACE inhibitors than men (RR 2.37, 95% CI 1.48 to 3.80, p =0.0003; RR 1.60, 95% CI 1.11 to 2.31, p=0.0131, respectively).

Predicting probability of drug use in OSAS patients

In order to determine whether any of the measured independent variables (age, BMI, AHI, time <90%, ESS) might predict the use of drugs for patients with OSAS, a logistic regression analysis was used with stepwise elimination. The results are summarised in table 5. Medication use could generally be predicted by age, sex, and ESS. The relationship between drug use and ESS is shown in fig 1. Cardiovascular drug use was predicted by age, BMI, and time<90%. Age, BMI, and AHI were predictors of antihypertensive drug use in patients with OSAS.

This analysis was performed for each antihypertensive drug. Table 6 shows that the use of diuretics and of ACE inhibitors was predicted by BMI and age. AHI was a predictor of diuretic and ACE inhibitor use and time<90% was a predictor of the use of loop diuretics. ESS did not predict the use of any cardiovascular drug.

Predicting cost of drugs in OSAS

Days of all drug supplied was significantly increased by age (p<0.0001), female sex (p<0.0001), BMI (p=0.0042), and time<90% (p=0.0005), but not AHI or ESS. The cost of all drugs supplied was also significantly increased by the same variables: age (p<0.0001), female sex (p<0.0034), BMI (p<0.0089), and time<90% (p=0.0012, fig 2).

Interestingly, for cardiovascular drugs the only variable which significantly predicted increased DDS (p<0.0001) and cost (p=0.0041) was age.

DISCUSSION

Our study has shown that patients with OSAS are heavy users of prescribed medications, especially those for cardiovascular disease, before their OSAS diagnosis and treatment. This was true both for men and women. Our findings are consistent with previous studies that have shown that hypertension is more often diagnosed in these patients. The relative risk for being prescribed an antihypertensive agent compared with controls (2.71) in our patients is of the same order of magnitude as the odds ratio for hypertension reported in recent large studies that employed very different experimental designs.^9,10 For example, Peppard et al found that the odds ratio of having hypertension in subjects with an AHI of >15 was 2.89 relative to those with an AHI of zero.¹⁰ However, our data on the use of antihypertensive agents must be interpreted cautiously since clinicians might use these medications not just for controlling high blood pressure, but for other indications and benefits from reducing blood pressure such as the treatment of angina pectoris, congestive heart failure and the secondary prevention of myocardial infarction in high risk patients. There is no way to determine from the database the specific indication for which a drug is prescribed.

Research has shown that sleep apnoea is an independent risk factor for the development of arterial hypertension. This research encompasses work in experimental animals,⁸ clinical studies,²³ epidemiological studies,^9,10 and health care utilisation studies.²⁴ The latter study showed that health care cost, as measured by utilisation of health care resources (physician fees), was higher for hypertension in undiagnosed OSAS patients than in controls.

Our study extends the earlier studies showing that the use of medications in these patients is increased compared with matched controls. Patients with apnoea who are not yet receiving treatment are therefore very heavy users of health care resources as measured by physician fees, hospital admissions, and medication use.

Cases and controls were matched as closely as possible for sex, age, and postal code using data from the population registry. The postal codes were matched because it has been shown that use of health care resources varies with socioeconomic factors including area of residence. The population database does not store data such as weight or height so it was not possible to match these patients by BMI. Patients and controls had been matched to the same family physician and therefore this helped to control for the fact that different physicians may use different criteria and medications in their practice.

Women with OSAS were more likely to be using medications than men, but not for cardiovascular drugs or antihypertensives. It has previously been reported that women are heavier users of drugs than men.¹⁶

Our study suggests that there are differences between men and women with OSAS, with apnoea severity (AHI) being lower and both age and BMI higher in women than in men. In the year before OSAS diagnosis women had been on more drugs and had been dispensed many more prescriptions at a much greater cost than men. The reasons for these sex differences are not known.

It was not surprising to find that age was a predictor for use and cost of all drugs and cardiovascular drugs. Drug use has been shown to increase with age.¹⁶ BMI did not predict the use of all medications, but did predict the use of cardiovascular drugs. The cost of all drugs was increased by sleep hypoxaemia. The cases with more severe apnoea (as measured by AHI) were more likely to be on antihypertensive agents—for example, patients with an AHI of 60 had a 65% increase in odds ratio of being on an antihypertensive agent compared with those with an AHI of 10. In contrast to the study by Nieto et al⁹ which found that sleep hypoxaemia increases the risk of hypertension, we found that exposure to hypoxaemia did not predict the use of antihypertensive drugs but did predict the use of loop diuretics. The latter medications are used most often to treat oedema. It has been suggested that hypoxaemia plays a role in the development of cor pulmonale and oedema in OSAS.²⁵

We found that subjective sleepiness in OSAS patients predicted the use of all medications but not the subgroup of drugs used to treat cardiovascular disease. This suggests that either patients are on non-cardiovascular medications acting on the nervous system that are making them sleepy, and/or that more sleepy patients are being treated for another condition such as depression. It has been suggested that features of depression may be present in OSAS and that patients with OSAS are more likely to have been diagnosed with depression.^26–28

This research has established the usefulness of the DPIN database in evaluating drug use of specific patient groups and for specific drugs. Future studies we will examine medication use in other OSAS co-morbidities such as depression and evaluate the impact on medication use of treatment in patients with OSAS.