A Time-Series Analysis of the Effect
of Increased Copayments on the Prescription
of Antidepressants, Anxiolytics, and Sedatives
in Sweden from 1990 to 1999
Michael Ong, MD, PhD, 1 Ralph Catalano, PhD, 2 and
Terry Hartig, PhD, MPH 3
1University of California, San Francisco, 2University of California, Berkeley, California, and
3Uppsala University, Uppsala, Sweden
Background: Outpatient prescription medication spending in Sweden has in-
creased sharply since 1974. The Swedish government has raised copayments to
reduce medication consumption and limit the growth of medication spending.
Objective: The aim of this study was to examine the effect of the 1995 and
1997 copayment increases on Swedish consumption of antidepressants, anxiolyt-
ics, and sedatives.
Methods: Monthly drug-use data for July 1990 through December 1999 for
these 3 pharmaceutical classes were obtained from Apoteket AB (Stockholm, Swe-
den). Data were provided for both sexes in units of defined daily doses per 1000
inhabitants. These series were analyzed with the use of Box-Jenkins autoregres-
sive, integrated, moving-average time-series modeling methods.
Results: Dispensing of all 3 drug classes increased immediately before copay-
ment changes, with the exception of male sedative use at the time of the 1997
reform. Permanent increases in male antidepressant and sedative use occurred be-
fore the 1995 copayment reform. Only female antidepressant use was perma-
nently reduced following the 1997 copayment reform.
Conclusions: Our findings suggest that Swedish patients' valuation of mental
health medications exceeds the enacted price increases. The permanent increases
in male antidepressant and sedative use, beginning in 1995, may have been the
result of previous undertreatment. The permanent reduction in female antide-
Accepted for publication February 3, 2003.
Published in the USA. Reproduction m whole or part is not permitted.
Copyright © 2003 Excerpta Nledica. Inc
M. Ong et al.
pressant use, beginning in 1997, suggests that the price levels reached a thresh-
old that matched or exceeded Swedish women's valuation of these medications.
(Clin Ther. 2003;25:1262-1275) Copyright © 2003 Excerpta Medica, Inc.
Key words: pharmacoeconomics, Sweden, mental health, copayments,
Outpatient pharmaceutical spending in Sweden has increased dramatically over
the past 3 decades. In 1974, total pharmacy drug sales were 1,816,000,000
Swedish kronor I (SEK) (in December 2002, 1 SEK = US $0.112). Pharmacy drug
sales reached 10,051,000,000 SEK in 1990 and 23,332,000,000 SEK in 1999.1
This value was >1% of the Swedish gross domestic product in 1999. 3 The in-
crease in drug sales led the Swedish government to institute cost-control mea-
sures to slow the growth of pharmaceutical expenditures. Sweden has chosen to
use copayments, requiring patients to assume a fraction of the actual cost. Co-
payments provide a means of reducing costs to the insurer while continuing to
provide insurance benefits to patients. 4 They may take the form of a deductible,
in which individuals pay a portion of the cost before insurance covers the re-
mainder, or coinsurance, in which individuals pay a fraction of the cost of each
Before 1997, Sweden used deductibles to limit pharmaceutical spending. These
deductibles slowly increased over time. The 15 SEK deductible established in
1974 rose to 40 SEK in 19836 and to 90 SEK in 1989. 7 In 1993, the government
increased the copayment to 120 SEK for the first prescription and reduced it to
10 SEK for prescriptions for additional drugs. 7 This deductible eventually in-
creased to 125 SEK for the first prescription and to 25 SEK for prescriptions for
additional drugs. In July 1995, copayment levels were raised to an initial pre-
scription cost of 160 SEK and 60 SEK for prescriptions for additional drugs, s
Each prescription allows a maximum 3-month supply of medication. 9
In January 1997, Sweden implemented copayment reforms raising the de-
ductible to 400 SEK, after which patients pay a proportion of the additional cost
up to a limit of 1300 SEK per year. This proportion varies according to the mag-
nitude of the total cost. 1° This reform represented a significant increase in phar-
maceutical costs faced by Swedish consumers.
The aim of this study was to examine the effect of these copayment increases
on Swedish consumption of antidepressants, anxiolytics, and sedatives. We hy-
pothesized that prescriptions for these drugs would decrease when patients were
made to pay a substantially higher fraction of the cost. We tested the hypothesis
separately for men and women because surveys of the Swedish population have
shown that women have a higher prevalence of depression, anxiety, and sleep dis-
orders compared with men.11
CLINICAL THERAPEUTICS c~
Previous research on Swedish demand for pharmaceuticals ~ has shown the
price elasticity, a measure of the sensitivity of indMduals to price changes, to be
about -0.1 to -0.3. These elasticity estimates suggest that for every 10% increase
in price, Swedish individuals will consume 1% to 3% less medication. The 1995
reforms increased the real price (the price after accounting for inflation t) of an
initial prescription by 24% and of subsequent prescriptions by 133%. Compari-
sons between the 1995 and 1997 reforms are more difficult to make because the
insurance plan was changed from requiring a copayment for each prescription to
requiring an initial deductible from each patient. In a hypothetical situation
where 2 prescriptions have a prereimbursement cost over 400 SEK, an individ-
ual would have paid 220 SEK in 1995 and 400 SEK in 1997 (and since 1992,
the average price of a prescription has been >200 SEK). 2 In real (ie, deflated)
prices, this is a 78% increase. Thus, a reduction in pharmaceutical consumption
seems logical, given the 1995 and 1997 reforms.
We focused on antidepressants, anxiolytics, and sedatives for several reasons, one
of which is that these compounds treat related illnesses. Depression and anxiety are
both mood disorders and insomnia is a common symptom of mood disorders. In-
clusion of other classes of medications would have reduced comparability.
In addition, these compounds represent a large fraction of the costs incurred
by the health insurance system. In 1999, Sweden spent 2,604,819 SEK (-17.4%
of all outpatient prescription drug spending) on mental health medications. 1 Only
drugs for endocrine and gastrointestinal disorders contributed more to outpatient
drug spending (-18%).
Another reason for our focus on these drugs is that they are used to treat ill-
nesses that put a significant burden on Swedish society. The prevalence of psy-
chiatric disorders in Sweden has been estimated at 14%. 11 Approximately 10% of
all primary care visits involve psychiatric diagnoses. 12 In 1999, 23.4% of all out-
patient prescriptions involved central nervous system agents--more than any
other category of outpatient prescriptions.1
We also focused on this group of drugs because patients and physicians in Swe-
den appear to view their use as more discretionary than that of other drugs. 13 We
hypothesized that these drugs would be used less often when patients faced
higher costs. We did not include antipsychotics in our analysis because these
drugs are unlikely to be considered discretionary.
We anticipated at least 2 complexities in the response of patients and physi-
cians to the copayment reforms. In the first, patients who take a medication on
a long-term basis might create a stockpile in anticipation of increased prices, with
the assistance of sympathetic physicians. In such a case, antidepressants are likely
to be stocked because they are more frequently used for extended periods of time
than sedatives or anxiolytics. In the second, in response to higher costs, there
might be an initial reduction in the use of antidepressants, anxiolytics, and seda-
M. Ong et al.
tives, followed by a rebound if clinicians or patients discovered they had under-
estimated the therapeutic value of the drugs.
We used the World Health Organization (WHO) Anatomical Therapeutic Classi-
fication system 14 to identify pharmaceuticals for this study. We chose antidepres-
sants (N06A), anxiolytics (N05B), and hypnotics and sedatives (N05C). Apoteket
AB (Stockholm, Sweden) provided drug use data for these categories. Data were
provided in units of defined daily doses (DDDs) per 1000 inhabitants. A DDD is
the assumed average maintenance dose per day for a drug used for its main in-
dication in adults. These DDDs are established by the WHO Collaborating Cen-
tre for Drug Statistics Methodology. 14 The DDD is a commonly used, standard-
dose concept that allows comparisons of prescription data over time. 15
The test period included the months from July 1990 through December 1999.
This span began 60 months before the July 1995 insurance reforms and ended
with the most recent data available at the time of our analysis. The DDD series
for antidepressants, anxiolytics, and sedatives in men and women are plotted in
Figures 1 and 2, respectively.
Randomized trials and quasi-experiments stipulate that an intervention has an ef-
fect only if the postintervention values of the dependent variable differ--after ac-
counting for known confounders--from those expected under the null hypothesis (ie,
under the assumption that the intervention had no effect). The randomized trial de-
rives the values expected under the null hypothesis from a non-self-selected control
group. The principal challenge in quasi-experiments is to derive those values through
another approach. Most quasi-experiments use statistical assumptions and manipula-
tions to derive the values expected under the null hypothesis. The simplest assump-
tion is that the mean of preintervention observations will be found postintervention
if the intervention had no effect. However, observations collected over time often ex-
hibit autocorrelation, including trends, seasonal and other cycles, and the tendency
to remain elevated or depressed after high or low values. Autocorrelation complicates
quasi-experiments because the expected value of an autocorrelated series is not its
mean. 16 The logic, 16 computational routines, 17 and decision rules 18 methods for de-
riving the values expected under the null hypothesis have been developed for auto-
correlated time-series data. Each method uses statistical manipulation to decompose
a dependent series into its autocorrelated and residual (unexpected) components. The
test then hinges on whether the residual components exhibit patterns consistent with
the hypothesis. The residual components support the hypothesis that an intervention
affected the dependent variable if they were above the 95% CI at the times specified
a priori as those in which the intervention should have an effect. 16
We used the method attributed to Box and Jenkins 19 to decompose our de-
pendent variables. The method--autoregressive, integrated, moving-average
CLINICAL THERAPEUTICS <~'
Figure I. Defined daily doses (DDDs), per 1000 inhabitants, of antidepressants, anxio-
lytics, and sedatives prescribed for Swedish men from July 1990 through De-
(ARIMA) time-series modeling--allows any of a large family of possible models
to be empirically fit to a time series.* The objective of the method is to determine
which combination of parameters in the following general ARIMA equation best
fits a time series:
(1 -- ~1 B2 -- ... -- d~pBp)VdZ t =
00 + (1 - 01B - 02 B2 - ... - OqBq)a t
~b = autoregressive parameter;
B = backshift operator that yields the value of the series it conditions at time t-p
for the autoregressive parameter or t-q for the moving average parameter (eg,
seasonality in a series would result in p or q values of 6 or 12, respectively);
*A technical appendix that describes the step-by-step application of these methods to our dependent variables
is available from R.C. (E-mail: email@example.com).
M. Ong et al.
I I I I
Figure 2. Defined daily doses (DDDs), per 1000 inhabitants, of antidepressants, anxio-
lytics, and sedatives prescribed for Swedish women from July 1990 through
V a = difference operator that indicates a series was differenced at lag d (ie, val-
ues at t subtracted from values at lag t-d) to remove secular trends or cy-
cles detected by the Dickey-Fuller test]7;
Z t = DDD per 1000 inhabitants of a given drug for month t;
0 o = mean (if any) of the DDD after differencing;
0 = moving average parameter;
a t = error term for week t.
We inspected the error terms from equation (1) for April 1995 through Octo-
ber 1995 and October 1996 through April 1997 for patterns consistent with ef-
fects of increased copayments. We used the method attributed to Demming 2° as
well as Alwan and Roberts 21 and described by Liu and Hudak. 22 Essentially, the
method searches for 3 types of changes indicated by different patterns of outliers.
We refer to these changes as steps, spikes, and decay. Steps are changes in which
the error terms move above or below the 99% (2-tailed test) CI of their expected
levels and remain, on average, outside the CI for the remainder ot7 the test period.
A spike is a change in which the error term for a single month is above or below
the 99% CI of the expected value of the error term. Decay alludes to a change in
which a spike decays geometrically such that at least 1 subsequent value remains
outside the 99% CI. A step would be specified in equation (1) as follows:
(1 -- q~l B -- ~)2 B2 -- -- ~pBP)VdZ, =
00 + col t + (1 "2" 01B _ 02B2 _ .
_ OqBq)a t
where I t is a binary variable scored 0 for all months before the initial outlier and
1 for all subsequent months, and co is the increase or decrease in Z associated
with intervention I.
A spike would be specified identically to a step but I t would be scored 0 for all
months before the outlier, 1 for the month of the outlier, and 0 afterward. Decay
would be specified as follows:
(1 - (~1B - ~2 B2 - ... - ~bpBP)VdZ t =
00 +l--'-~It + (1 - 01B - e2 B2 - ... - OqBq)a t
where I t is a binary variable scored 0 for all months before the outlier, 1 for the
month of the outlier, and 0 afterward, and 8 is the proportion of m that is car-
ried into the next month.
We also used the methods described above to search for and control outliers
in months other than those in which the copayment policy implies that we would
find outliers. Including these outliers in the models should reduce variance in the
error terms that could otherwise lead to false acceptance of the null hypothesis
for the effects of copayments.
Tables I and II show the best fitting ARIMA models and (as described in the
"Methods" section) identify patterns of change in the times most likely affected
by the 2 increases in copayments. Only significant outliers are shown (ie, P < 0.01
or P < 0.05; 2-tailed test). The only permanent changes were increases in male
antidepressant and sedative DDDs occurring prior to the 1995 reform and a re-
duction in female antidepressant DDDs occurring after the 1997 reform (which
was slightly tempered 3 months afterward).
The ARIMA models imply that, as expected from Figures 1 and 2, the dis-
pensing of antidepressants and anxiolytics to men and women rose over the
course of the test period. The fact that every series exhibited seasonal cycles may
Table I. Autoregressive, integrated, moving-average (ARIMA) time-series equations and outlying values for prescriptions for anti-
depressants, anxiolytics, and sedatives dispensed to Swedish men from July 1990 through December 1999.
Outliers from April 1995
Outliers from October 1996
through October 1995
through April 1997
VZ t -- 0.216" + (I - 0.652"B)(I + 0.378"Bt2)a t
May 1995 step -- 2.740t
November 1996 spike = 4.6745
July 1995 decay = -5.275t
December 1996 decay = 10.474t
January 1997 decay = -4.393 t
- 11.253") = a,
(i - 0.765"B)(i
June 1995 spike = 1.759~
May 1995 step -- 3.477t
July 1995 decay = -5.838 t
January 1997 decay = -I.600t
November 1996 spike -- 6.382t
December 1996 decay -- 7,703t
January 1997 decay -- -3.415t
"P < 0,05, 2-tailed test.
tp < 0.0 I, 2-tailed test.
Table II. Autoregressive, integrated, moving-average (ARIMA) time-series equations and outlying values for prescriptions for anti-
depressants, anxiolytics, and sedatives dispensed to Swedish women from July 1990 through December 1999.
Outliers from April 1995
Outliers from October 1996
through October 1995
through April 1997
(I -0.369*B~2)VZ~ = (I - 0.516"B)o t
June 1995 spike = 6.618t
November 1996 decay = 9,079t
December 1996 decay
January 1997 step = 21.129t
April 1997 step -- 4.460i
(~ - o.soo'B,~)vz,
= (i - 0.798"B)a,
June 1995 spike -- 2.861 t
June 1995 spike -- 8.734f
December 1996 decay -- 2.739:
January 1997 decay -- 3,548-
December 1996 decay -- 12.201,
January 1997 spike = -I 1,304f
*P < 0.05, 2-tailed test.
tp < 0.01,2-tailed test
H. Ong et al.
be less obvious to the eye, but is clear from the Box-Jenkins analyses. These cy-
cles were controlled by the moving average and autoregressive parameters at lags
6 and 12.
Men's use of all 3 drugs increased in the 2 months immediately preceding the
1995 reforms (ie, May and June), Dispensing of antidepressants and sedatives to
men stepped up in May 1995 and were elevated through the remainder of the
study period. Prescriptions for anxiolytics temporarily increased in June 1995.
Dispensing of both antidepressants and sedatives to men was lower during the
month that the higher copayments took effect (ie, July) but decayed back to the
levels observed in May.
The dispensing of all 3 drug classes to women spiked upward in the month be-
fore the implementation of increased copayments in 1995 (ie, June).
Around the 1997 reforms, dispensing of both antidepressants and sedatives to
men spiked in November and increased in December but decayed in subsequent
months. The decay of these increases was accelerated by decaying decreases in the
month that the copayrnents rose (ie, January 1997). Dispensing of anxiolytics to
men dropped in January and subsequently decayed back to the expected level.
The January 1997 increase in copayments followed December increases in the
dispensing of all 3 drugs to women. All increases subsequently decayed. The De-
cember increases preceded temporary January reductions for both anxiolytics and
sedatives. The January reduction in dispensing of anxiolytics decayed back to ex-
pected levels, whereas that for sedatives was a single downward spike.
The decaying December increase in antidepressants followed a smaller but de-
caying increase in November. As with anxiolytics and sedatives, the quantity of
antidepressants dispensed to women dropped in the month in which copayments
increased (ie, January 1997). However, the January drop for antidepressants ap-
peared to be permanent. A smaller step upward in April 1997 adjusted the per-
manent level of the January drop.
We estimated our equations again after making the distributions of the depen-
dent variables more normal through natural logarithm transformations. The
transformations did not change the results of the test.
Economic theory and previous research suggest that pharmaceutical consumption
should decrease as consumer costs increase. However, our analyses produced lit-
tle support for this outcome. Only antidepressant use among women was per-
manently reduced after the 1997 reforms. All other medication classes were per-
manently increased or experienced temporary fluctuations in use for both sexes.
The most likely explanation for our findings is that Swedish patients are rela-
tively insensitive to these price increases. There are several possible causes of con-
sumer price insensitivity: (1) even after price increases, the cost to consumers
CLINICAL THERAPEUTICS C ~
may still be significantly below what they are willing to pay; (2) there may be no
available substitutes; and (3) the price increases may be insignificant. If any of
these scenarios existed, it would provide a reasonable explanation for our find-
ings. Here, we consider each of these cases in turn.
In the first scenario, high valuation of mental health pharmaceuticals could ex-
plain the apparent lack of consumption decline with raised prices. If the current
prices were much lower than Swedish patients were willing to pay for these 3
classes of medications, an increase in prices that did not approach the Swedish
willingness-to-pay limit would result in no behavior change. This explanation is
supported by the low dollar value of Swedish copayments and deductibles (even
after the price reforms), which may well be below the level at which most Swedish
patients would be willing to pay
In the second scenario, patients might elect to pay higher prices for medications
if substitute goods were not available. For example, if psychotherapy treatments
were relatively inaccessible (and therefore expensive), consumers would not switch
from pharmaceutical therapy to psychotherapy unless drug prices were high enough
to make both treatment types equal in expense. Outpatient mental health treatment
data for 1997 show that about 14% of outpatient visits were made to psychother-
apists and psychologists, but together these 2 professions averaged about 2.3 visits
per provider. 23 This suggests that pharmaceutical substitutes are accessible.
In the third scenario, negligible price changes would be unlikely to change the
consumers' perception of the value of pharmaceutical treatments. However, this the-
ory is less likely than the others to explain our findings; stockpiling of drugs---in
which the quantities of drugs dispensed increased at the time of or immediately pre-
ceding a price increase--was observed for all 3 drug classes studied within both pe-
riods of price changes, with the exception of anxiolytic use among men during the
1997 reform. Unfortunately, our data were not specific enough to determine whether
these temporary increases in consumption were due to existing users or new users.
Permanent increases were seen in antidepressant and sedative dispensing
among men beginning in 1995. Because our data are limited to DDDs, it is un-
clear whether these increases reflect increases in medication units used or reduc-
tions in the population using the drugs. The latter explanation is unlikely, given
that the Swedish population increased every year during the study period) In-
creases in medication unit use could be the result of increases in either the num-
ber of users or total doses dispensed. An increased number of users would sug-
gest increased diagnosis of disease, increased trials of therapy, or increased
incidence of disease. Mental illness is generally underdiagnosed and undertreated
globally. 24 However, it is unlikely that improved diagnosis would occur in 2 ill-
nesses during the period immediately preceding a price increase.
A more plausible explanation would be that the impending price increases may
have spurred patients to seek treatment and physicians to attempt trial of ther-
M. Ong et al.
apy before a rise in cost. Individuals benefiting from treatment would likely con-
tinue therapy, resulting in a permanent increase in the use of the drug. Such be-
havior could mask any reductions in consumption among prior users due to in-
creased costs. The permanent increases in antidepressant and sedative use among
men during the 1995 reforms may be the result of empirical trials of treatment
that persisted. We do not know of any other phenomenon occurring during this
period that might have reduced undertreatment of depression and insomnia in
men. Increased incidence of mental illness could also increase medication use.
However, concomitant rises in medication use and in pharmaceutical prices
should only occur if the price increases caused a rise in diagnosis of mental ill-
ness or if some other phenomenon occurred at the same time. Because we did
not find such rises in medication use in conjunction with larger price increases
in 1997, it seems unlikely that price increases caused greater incidence of diag-
nosis of mental illnesses. In addition, we know of no other phenomenon that co-
incided with the price increases. Increases in total dose are also an unlikely ex-
planation. There were no changes in DDD definitions in any of the drug classes
during the period studied. There were also no widespread phenomena that might
have increased the severity of depression or insomnia in Swedish men or led to
increased doses for existing users.
Why was there a reduction in antidepressant DDDs among women in 1997?
Based on our previous explanations for the lack of response in other categories,
we believe that the threshold at which Swedish women were willing to pay for
antidepressants may have been reached. Given our observation of an effect on
prescriptions in this population, it would be difficult to argue that the price in-
creases were negligible. In addition, the absence of an observed inhibitory effect
on antidepressant prescriptions for men suggests that availability of medication
substitutes may not have been an issue in Sweden overall during the time period
The aforementioned absence of an inhibitory effect on antidepressant prescrip-
tions among men may be the result of lower income for women than men, 3 which
might make medication prices relatively higher for women compared with men.
Women may have reached the threshold price level at which their antidepressant
use declines with subsequent price increases.
Our analysis is limited by the lack of specificity in DDD reported data. Al-
though the DDD methodology allowed for standardized comparison across dif-
ferent pharmaceutical classes, we were unable to differentiate between specific
medications. There may have been differences between drugs within each phar-
maceutical classification. For example, serotonin-selective reuptake inhibitors and
tricyclic compounds are both antidepressants, but consumer demand for them
may have varied over time. Unfortunately, our data were not specific enough to
overcome the problem of heterogeneity. Further work in this area is warranted.
CLINICAL THERAPEUTICS ®
Although no permanent declines in drug use tollowing copayment reforms
were identified in this study, the Swedish government accrued benefits with these
policy changes. The burden of pharmaceutical spending has been shifted away
from state budgets onto patients who use and benefit from these therapies. Fur-
ther studies are needed to determine whether our findings could be applied to
drugs other than those used for mental health care or to countries other than
Sweden. The external validity of this study would be improved by comparing
these findings with the effects of copayment increases on prescriptions in another
country. Such future comparisons are warranted.
We did not find widespread reductions in mental health drug use after the im-
plementation of copayment increases in Sweden in I995 and 1997. However, we
observed a permanent reduction in women's use of antidepressants after the 1997
copayment reform. In addition, we found permanent increases in men's use of an-
tidepressants and sedatives after the 1995 copayment reform, despite the elimi-
nation of trends from the data using ARIMA modeling techniques. These findings
suggest that patients may value these medications more highly than was indicated
by previous consumption levels. The 1995 increases observed in men's antide-
pressant and sedative use may have been the result of previous undertreatment.
The permanent reduction observed in women's antidepressant use suggests that
the price levels reached a threshold that matched or exceeded the patients' valu-
ation of these medications.
1. Svensk Lakemedelsstatistik 1999. Stockholm, Sweden: Apoteket AB; 2000.
2. The Universal Currency Converter Web site. Available at: www.xe.com. Accessed De-
cember 1, 2002.
3. Statistics Sweden Web site. Available at: http://www.scb.se/indexing.asp. Accessed
4. Leibowitz A, Manning W, Newhouse J. The demand for prescription drugs as a func-
tion of cost sharing. Soc Sci Med. 1985;21:1063-1069.
5. Feldstein PJ. Health Care Economics. 5th ed. Albany, NY: Delmar Publishers; 1999.
6. Wessling A. Continuous recording of drug prescribing in Sweden 1974-1983. Meth-
ods and examples of utilization of data. EurJ Clin Pharmacol. 1987;33:7-13.
7. Gross DJ, Ramer J, PerezJ, Glavin SL. International pharmaceutical spending controls: France,
Germany, Sweden, and the United Kingdom. Health Care Financ ~
8. Ljungkvist M, Andersson D, Gunnarsson B. Cost and utilisation of pharmaceuticals
in Sweden. Health Policy. 1997;41(Suppl):S55-S69.
9. Jonsson B. Cost sharing for pharmaceuticals--the Swedish reimbursement system.
PharmacoEconomics. 1996;10(Suppl 2):68-74.
M. Ong et al.
10. Henriksson E Hjortsberg C, Rehnberg C. Pharmaceutical expenditure in Sweden.
Health Policy. 1999;47:125-144.
11. Sweden's Public Health Report 1997. Stockholm, Sweden: Socialstyrelsen, Moden-Tryck
12. Stefansson CG, Svensson C. Identified and unidentified mental illness in primary
health care--social characteristics, medical measures and total care utilization during
one year. Scand J Prim Health Care. 1994;12:24-31.
13. Lindgren B, Silverberg R. Subsidizing pharmaceutical prescriptions: The experience
of Denmark, Finland, Norway and Sweden. Soc Sci Med. 1985;21:1089-1096.
14. WHO Collaborating Centre for Drug Statistics Methodology Web site. Available at:
http://www.whocc.nmd.no. Accessed February 2001.
15. Merlo J, Wessling A, Melander A. Comparison of dose standard units for drug utili-
sation studies. Eur J Clin Pharmacol. 1996;50:27-30.
16. Catalano R, Serxner S. Time-series designs of potential interest to epidemiologists. Am
J Epidemiol. 1987;126:724-731.
17. Hamilton J. Time Series Analysis. Princeton, NJ: Princeton University Press; 1994.
18. Ljung GM, Box GE. On a measure of lack of fit in time series models. Biometrika.
19. Box GE, Jenkins GM, Reinsel GC. Time Series Analysis: Forecasting and Control. 3rd ed.
Englewood Cliffs, NJ: Prentice Hall; 1994.
20. Demming WE. Quality, Productivity and Competitive Position. Cambridge, Mass: Mas-
sachusetts Institute of Technology, Center for Advanced Engineering Study; 1982.
21. Alwan L, Roberts H. Time-series modeling for statistical process control. J Bus Econ
22. Liu L, Hudak G. Forecasting and Time Series Analysis Using the SCA Statistical System.
Oak Brook, Ill: Scientific Computing Associates; 1992.
23. Silfverhielm H, Kamis-Gould E. The Swedish mental health system. Past, present, and
future. Int J Law Psychiatry. 2000;23:293-307.
24. US Department of Health and Human Services. Mental Health: A Report of the Surgeon
General. Rockville, Md: US Department of Health and Human Services, Substance
Abuse and Mental Health Services Administration, Center for Mental Health Services,
National Institutes of Health, National Institute of Mental Health; 1999.
Address correspondence to: Michael Ong, MD, PhD, University of California, San
Francisco, Division of General Internal Medicine, Ambulatory Care Center, 400
Parnassus Avenue, Box 0320, San Francisco, CA 94143-0320. E-mail: mong@