Antidepressant treatment patterns and costs among US employees

Abstract

Objective: To compare medical and cost profiles of patients treated for depression classified by treatment pattern groups.

Methods: An analysis used de-identified 1999–2004 employer claims data (n=2.9 million beneficiaries) for employees with ≥1 diagnosis of major depressive disorder and ≥1 antidepressant prescription, following a 6-month washout period of no antidepressant prescription. Patients were classified into switcher/discontinuer/augmenter/maintainer during Healthcare Effectiveness Data and Information Set-defined initial and subsequent treatment periods, then grouped into stable, intermediate, or non-stable treatment groups, based on stability of treatment patterns. Medical/cost profiles for 6-month pre- and 12-month post-index periods were compared descriptively, and multivariate regressions were estimated, controlling for baseline characteristics/severity markers. Cost savings reflect differences between treatment pattern groups from a current US perspective.

Results: Of the 5,225 patients meeting inclusion criteria, 60.8% were in stable, 24.5% in intermediate and 14.7% in non-stable treatment groups. No significant differences existed in medical profiles and costs between the three groups in the pre-index period. In the post-index period, stable group patients had lower costs compared to intermediate and non-stable groups. Stable group patients generated cost savings of $1,842 compared to intermediate and $5,231 compared to non-stable groups. Multivariate analysis confirmed these findings.

Conclusion: Patients on a more stable treatment regimen yield significant cost savings compared to patients on a less stable regimen.

Key words::

• antidepressants
• cost
• depression
• treatment patterns

Introduction

Major depressive disorder (MDD) affected 18.1 million people in 2000 in the US, and in 2005, MDD had a lifetime prevalence of 16.2%^1,2. Patients with depression tend to have more comorbid conditions than those without MDD, particularly other mental illnesses and chronic diseases³.

The economic consequences of MDD are substantial because treating the disease itself, as well as its accompanying comorbidities, requires substantial healthcare resources^1,4. Moreover, significant indirect costs result from suicides and suicide attempts, as well as reduced productivity both at paid and unpaid work, and increased disability associated with the disease^1,4. The World Health Organization projected that major depression will become the second leading cause of disability worldwide by 2020⁵.

Greenberg et al estimated the annual societal costs of clinical depression at $83.1 billion in 2000, an increase of 7% since 1990 after inflation adjustment^1,4. Because the treatment rate of depression increased by over 50% and costs remained relatively stable, the average cost of treatment per patient decreased. This was the result of a switch away from hospitalisation as well as a reduced role for care by psychiatrists in favour of an increased role for primary care physicians and greater use of prescription pharmaceuticals. Despite broad use of antidepressants, Kessler et al estimated that MDD still resulted in 27.2 lost workdays per ill worker each year⁶. The literature on the economic consequences of treatment failure for patients with depression indicates that the costs could be quite substantial^7–10.

Among the depressed population on medication, Thompson et al found that the costs of care were highest among patients who discontinued early, switched, or added treatment¹¹. This is noteworthy because 40–50% of primary care patients diagnosed with depression discontinue treatment within the first 3 months¹². In addition, patients who discontinue early are more likely to experience a depression relapse^13,14. These studies implied that different treatment patterns can result in varying rates of remission and relapse. Logically, patients with treatment-resistant depression (TRD) are more likely to switch and/or add treatments. As a result, patients with TRD will incur additional costs among the overall depressed population, if only for their extra medications. These results were confirmed in a claims data study that compared the costs between TRD patients, MDD patients and a control group. The study found that while costs are higher for MDD patients than for the control group, when MDD patients were split into two sub-groups, TRD likely and TRD unlikely, the annual per capita costs of TRD likely patients were twice those of the TRD unlikely group⁷. In other words, TRD patients incurred most of the cost difference when comparing MDD patients with the control group.

The main treatment modalities for MDD include antidepressant medication, psychotherapy and electroconvulsive treatment. To treat depression effectively, the Agency for Healthcare Research and Quality and medical specialty groups developed the Healthcare Effectiveness Data and Information Set (HEDIS) treatment guidelines to measure antidepressant management, specifying an 84-day acute treatment phase and an overall 180-day continuation treatment phase¹⁵. Akincigil et al examined adherence to antidepressant treatment using the HEDIS-defined phases and found that 51% of patients adhered to treatment through the acute phase and 42% continued through the continuation phase¹⁶. Burton et al examined the relationship between antidepressant treatment adherence with employee short-term disability absences and found that employees who were not adherent to HEDIS acute treatment guidelines were 38.7% more likely to have short-term disability claims¹⁷. The study reported significantly higher incidences of short-term disability during both the acute and continuation phases among non-adherent patients.

The purpose of the current research is to develop categories of treatment patterns and to compare direct (medical and prescription drug) and indirect (work-loss) costs of those treatment patterns. Because claims data lack clinical detail and it can be difficult to categorise treatment patterns, we seek to use claims data to identify clear instances of stable and unstable treatment, and then to identify ‘grey’ areas where it is not clear whether patients achieve symptom management or not. In other words, we try to identify clear patterns that imply stable and non-stable treatment and categorise the residual as ‘intermediate’. By developing a methodology to categorise patients in terms of stability in treatment based on a patient's specific treatment pattern and how it changes during the acute and continuation treatment phases, this economic study evaluates, from a current US perspective, the extent to which patients with MDD on stable treatment regimens incur lower direct and indirect work-loss costs than those on non-stable treatment regimens.

Patients and methods

Data source

The data source was a de-identified administrative claims database of 16 US employers' privately insured claims (1999–2004) covering 2.9 million beneficiaries. Although not intended to be a statistically valid representation of the national population, these employers have operations nationwide in a broad array of industries and job classifications, and are geographically diverse. The data contain de-identified information on patients' demographics (e.g. age and gender), medical and pharmacy claims, and monthly eligibility files. Specifically, patients' utilisation of medical services was recorded with date of service, actual payments to providers, associated diagnoses (using the codes for International Statistical Classification of Diseases and Related Health Problems, ICD–9), performed procedures (Current Procedural Terminology), etc. Patients' pharmacy claims contain prescribed medications recorded with National Drug Code, dates of prescription fills, days of supply, strength, quantity and the actual amount of payments to pharmacies.

Sample selection and study cohorts

Using the administrative claims database, analysis was restricted to employees, aged 18–64, who had at least one diagnosis of MDD (ICD–9: 296.2x, 296.3x) at any time in their claims history and at least one prescription filled for certain antidepressants (specifically, for selective serotonin reuptake inhibitor, serotonin/norepinephrine reuptake inhibitor, or bupropion) following a 6-month washout period of no antidepressant prescriptions. The initial antidepressant therapy was defined as the index drug. The date of therapy initiation was defined as the index date. Patients were continuously enrolled in the 6-month pre-index and 12-month post-index date periods. Table 1 summarises the remaining sample count after applying each inclusion and exclusion criterion.

Treatment patterns were observed during a 180-day period. Two time periods – an initial and a subsequent period – were defined after the index date. The initial period corresponding to the 12-week HEDIS ‘acute phase’ started on the index date and ended on day 84¹⁵. The subsequent period started on day 85 and ended on day 180. Patients were first classified into treatment pattern groups based on treatment events and changes in their antidepressant treatment pattern over the initial and subsequent periods. During the initial period, based on the first treatment event following the index date, patients were classified into one of four mutually exclusive treatment patterns: switcher, add-on, discontinuer and maintainer. Switchers ‘switched’ from the index drug to another antidepressant, without a refill of the index drug, and within a 45-day period after the end date of the index drug prescription. Add-on patients filled an additional antidepressant drug prescription on or after the index date, in conjunction with continuing refills of the index drug, and with prescriptions of the index drug no more than 45 days apart. Discontinuers did not switch or add treatments, and had no refill of the index drug within 45 days after the end date of the index drug prescription. Maintainers continuously refilled the index drug, with prescriptions no more than 45 days apart.

Similarly, during the subsequent period, based on the first treatment event following the first day of the subsequent period or day 85, patients were again classified into one of the four mutually exclusive treatment patterns described above or into a fifth ‘no-treatment’ group. ‘No-treatment’ patients had no antidepressant prescription filled during the subsequent period.

By combining the four treatment patterns of the initial period and the five of the subsequent period, each patient was categorised into one of 20 mutually exclusive treatment pattern groups. Figure 1 illustrates some examples. For instance, if a patient was classified as a switcher both in the initial and the subsequent periods, he/she was considered a ‘switcher-switcher’.

Using these treatment pattern groups, MDD patients were then stratified into three mutually exclusive treatment pattern groups – non-stable, stable and intermediate – based on evidence of stability in treatment therapy. Because switching and adding treatments are evidence of instability during the time period in which the pattern is observed, ‘non-stable’ patients either switch or add on treatments during the subsequent period and do not achieve stability of symptom management. In contrast, long-term maintenance and eventual independence from treatment are evidence of stability. Hence, ‘stable’ patients maintain treatment during the initial period without switching or adding treatment during the subsequent period. Alternatively, patients in the stable group discontinue treatment during the initial period without restarting another treatment in the subsequent period, and hence they do not require further treatment. All other patients who switch or add treatment only during the initial period or discontinue treatment during the initial period and later restart another treatment during the subsequent period are considered ‘intermediate.’

Measures

Direct (medical and prescription drug) and indirect work-loss (disability and absenteeism) costs were analysed during the 6-month pre-index period and the 12-month post-index period. Medical costs were calculated based on reimbursements from the payer to healthcare providers for inpatient, hospital outpatient (e.g. outpatient surgery), physician services, emergency department visits, as well as for other ancillary services (e.g. physical therapy, laboratory services). Inpatient costs were defined as claims with a place of service specified as hospital inpatient, rehabilitation centre, residential treatment centre, or psychiatric facility. All other medical costs were grouped into an outpatient and other costs category. Medical and drug costs were converted to 2005 US dollars using the medical services component of the Consumer Price Index.

Indirect work-loss costs include disability costs (i.e. employee compensation often in the range of 60–80% of wages) paid by employers for 6 or more workdays missed due to medical reasons) and medically related absenteeism (non-disability) of less than 6 days based on actual wages. For each day in which an employee had at least one medical claim, a day of medically related work loss was counted for a hospital day, and half a day of medically related work-loss was counted for other services. Indirect work-loss costs were converted to 2005 US dollars using the general consumer component of the Consumer Price Index.

Study sample characteristics and baseline comorbidities in the 6-month pre-index period were described for treated non-depression mental health and physical comorbidities and other selected measures. Non-depression mental health comorbidities included generalised anxiety disorder, panic disorder, post-traumatic stress disorder, phobia, obsessive-compulsive disorder, sleeping disorder, bipolar, other psychotic disorders, eating disorder and schizophrenia. Physical comorbidities listed in the Deyo-Charlson Comorbidity Index were used¹⁸. Other measures included substance abuse diagnosed, injuries and accidents and urgent care, as well as age and gender.

Statistical analyses

All outcomes were first compared descriptively among the treatment pattern groups in the 6-month pre- and 12-month post-index date periods using pair-wise comparisons. Cost variables were compared using ANOVA and post-ANOVA Bonferroni comparison. The Bonferroni comparison was conducted in order to protect against type I errors as multiple comparisons were being performed simultaneously. Three mutually exclusive treatment pattern groups were analysed, and a total of three simultaneous comparisons were required for each outcome. Therefore, under the Bonferroni correction assumption, significance was drawn only when the p-value was below 0.05/3 = 0.016.

A two-part regression model was then used to further control for potential differences in baseline characteristics (i.e. age, gender, index therapy, baseline comorbidities) and severity markers (i.e. prior hospitalisations, emergency department visits and diagnosis of substance abuse) across treatment pattern groups. A two-part model was used to estimate direct (medical and drug) and indirect (medical absenteeism and disability) costs because some patients had no claims for some or all services and hence had zero costs. More specifically, the first part was a logistic regression predicting whether any costs were incurred, and the second part was a generalised linear model (GLM) with log link and gamma distribution to determine the cost amount. These two parts generated the predictive models for costs for all patients.

A two-part regression model is a typical approach when the data being analysed have many zero values and at the same time a number of very large values. This scenario is typical in medical expense datasets, and hence the two-part regression model is an ideal approach¹⁹. Ignoring the special distribution of such data would lead to unreliable estimates. When a classic regression (such as ordinary least square [OLS]) is performed, only responses with non-zero values are taken into account, and all relationships between responses with zero values and its covariates are lost. In contrast, the two-part regression model uses all observations (both zero and non-zero) and ensures unbiased estimates towards non-positive values.

Seven separate models were estimated: one for total (direct plus indirect) cost, and for each cost category, direct and indirect costs, as well as their components: medical and drug costs; and medical absenteeism and disability costs. For each cost component, the two-part model first estimated the probability of costs being positive with logistic regression, and given positive costs, then estimated the predicted costs using a GLM with gamma distribution of the error term and log link function. This model was used to resolve the issue of a skewed cost distribution common in claims data analysis²⁰. In contrast with the traditional log-ordinary least squares regression (log-OLS), a GLM provides more robust coefficient estimates²¹. Also, the log link function of the mean response enables coefficients to be directly back transformed into the original dollar scale and avoids the issue of potentially biased estimates that may result from using the Duan smearing estimation method²². Variance for 95% confidence intervals was obtained using bootstrapping and differences were assessed by t-tests.

Results

The final study sample included a total of 5,225 employees who met the study criteria. Table 1 presents the sample count for all inclusion and exclusion criteria. Table 2 presents the distribution of patients by treatment patterns in the initial and subsequent periods and by treatment groups. Among these patients, 3,177 (60.8%) were in the stable treatment group, 1,279 (24.5%) were in the intermediate treatment group and 769 (14.7%) were in the non-stable treatment group. Baseline characteristics are reported by treatment pattern group in Table 3. Results show that patients shared similar demographics across treatment groups, had an average age of 41, and 36–39% were male. Patients also had similar mental and physical comorbidities, diagnoses of substance abuse and accidents/injury.

During the 6-month pre-index period, direct costs computed using descriptive analyses were not significantly different across treatment groups. Total direct costs were $2,572 for the stable group, $2,952 for the intermediate group and $2,825 for the non-stable group (Figure 2). Both medical and drug costs were statistically similar across the three groups. Similarly, the work-loss costs for the stable ($912), intermediate ($943) and non-stable ($949) groups were not statistically different. This was also true for both disability and absenteeism costs. The 6-month pre-index baseline characteristics were also statistically similar across treatment groups (Table 3).

During the 12-month post-index period, costs increased among all treatment pattern groups and differences in costs became statistically significant. Total direct costs computed using descriptive analysis were significantly lower in the stable group ($6,064) than in the intermediate ($7,317) and non-stable ($9,467) groups (Figure 3). Similarly, the medical costs were lower for the stable group compared to the other groups. The drug costs for the stable group ($1,649) were higher than those for the intermediate group ($1,406) because the stable group generally maintained their treatment longer than the intermediate group, while the non-stable group had the highest drug costs ($2,482). The total indirect work-loss costs were significantly lower in the stable group ($2,189) than in the intermediate ($2,776) or the non-stable ($4,019) groups (all p<0.001). This trend was seen in disability and absenteeism costs as well as in total direct and indirect costs.

Controlling for baseline characteristics and severity markers, results remained statistically lower for patients in the stable group and very comparable to the dollar amounts estimated in the descriptive analyses (Table 4). Using the cost estimates from the multivariate analysis, cost differences between the treatment pattern groups were calculated. Patients in the stable group incurred $1,102 lower direct costs and $608 lower indirect costs compared to patients in the intermediate group. Similarly, patients in the stable group incurred $3,732 lower direct costs and $1,904 lower indirect costs compared to patients in the non-stable group (all p<0.001).

Discussion

Patients with MDD can be categorised into stable, intermediate and non-stable treatment pattern groups based on their patterns of use of antidepressants. The stable group, which includes patients with first-line treatments, is the predominant group and accounts for 61% of the sample, while the intermediate, which includes patients with second-line treatments, accounts for 25% of the sample. This distribution is consistent with results from the ‘Sequenced Treatment Alternatives to Relieve Depression’ (STAR*D) study, where the response rate was 47%²³ and 31% had remission at second-line treatments²⁴. Note that this study is not an effectiveness study and does not assume that unstable treatment patterns represent the consequences of less effective treatment. Rather, we are describing real-world treatment patterns and the observed association with costs.

During the 6-month pre-index period, there was no statistical difference in direct and indirect work-loss costs between the treatment pattern groups. After initiating treatment with antidepressant therapy, costs increased among all treatment pattern groups and differences in costs among groups became significant. During the 12-month post-index period, compared to patients of the intermediate and non-stable groups, MDD patients of the stable group had lower direct and indirect work-loss costs. Because stable patients include mostly maintainers and intermediate patients include mostly discontinuers, the intermediate patients incurred lower drug costs than stable patients. Multivariate models controlling for baseline characteristics and severity markers validated the descriptive analysis of direct and indirect work-loss costs.

Despite greater drug costs among patients of the stable group, these patients incurred greater cost savings when medical and indirect work-loss costs are considered. The continuity of care, which is represented by the stable group, is associated with significant cost savings. Stable patients had total cost savings of $1,842 compared to intermediate patients and $5,231 compared to non-stable patients. Intermediate patients had cost savings of $3,389 compared to non-stable patients. Indirect work-loss costs accounted for approximately 35% of total cost savings and hospitalisation costs accounted for approximately 7% and 11% of the total cost savings, and direct cost savings, respectively. Multivariate analyses confirmed the magnitude of these cost savings. Note that much of the workplace burden of depression for employees lies beyond the measures used in this study – specifically at-work productivity loss. Indeed, research by Kessler suggests that the major workplace impact of depression is likely to be captured as presenteeism, which can be several times the cost of work-loss time alone⁶.

Another perspective on treatment groups is found in the treatment resistant depression literature. Different treatment patterns can be the result of TRD. A study conducted by Russell et al found that depression-related costs increased with the severity of TRD⁹. Similar results were found by Corey-Lisle⁸ et al, as well as Greenberg et al¹⁰. While research reported in this study only considers treatment patterns of antidepressant use, the non-stable treatment group may be clinically similar to treatment resistant depression patients, as these patients require treatment switches or addition therapies¹⁰.

Limitations

This study has several limitations. First, claims data do not permit the assessment of direct measures of disease severity or clinical outcomes and do not provide reasons for observed treatment patterns (i.e. switch, add-on, discontinue, maintain and no-treatment). Indeed, for a variety of misclassification reasons, not all patients treated for depression are identifiable by ICD–9 code classification. Therefore there may be selection bias if patients in one of the treatment groups had more severe depression than patients in another group, which may have affected treatment compliance and healthcare utilisation during the study period. However, every effort was taken to control for baseline characteristics, including multiple comorbidities and baseline healthcare utilisation, and results show no significant difference between treatment groups at baseline. Therefore, selection bias, if any, has been minimised in this study. Future research could consider medical costs associated with ailments related to prolonged stress associated with depression (i.e. stress leading to increased serum cortisol leading to insulin resistance and diabetes), obesity, cardiovascular disease, etc. A second limitation is that, as with all other claims data studies, ICD–9–CM coding could have affected the outcomes of this study through inclusion of non-MDD patients. Alternatively, some patients may request from their physicians that they do not code them as having depression because of the stigma of mental illness. A third limitation is that results underestimate indirect work-loss costs because they do not capture presenteeism and sporadic absenteeism. A fourth limitation is that this study did not attempt to look at the implications of subtypes of depression or comorbidities (e.g. depression with anxiety) on costs. The determinants of the cost of depression treatment were analysed in a separate study, where we showed that patients with MDD and comorbid generalised anxiety disorder or any type of anxiety had significantly higher costs than MDD-only patients²⁵. A fifth limitation is that this study relied on a sample of employed MDD patients in the US, and results are not meant to be representative of all MDD patients. Finally, treatment pattern categories are based on the first treatment event following the index date or day 85, and therefore do not reflect multiple treatment events during the initial or subsequent periods.

Conclusions

Results from this claims data analysis indicate that patients with MDD can be categorised into stable, intermediate and non-stable treatment groups based on their patterns of use of antidepressants and that non-stability of treatment yields statistically significant increases in direct and indirect work-loss costs. For non-stable MDD patients who require additional treatment therapies, better effectiveness in drug therapy has the potential to substantially increase cost savings.

Figure 1. Examples of treatment pattern groups in the initial and subsequent periods.

Figure 2. Direct and indirect work-loss costs during the 6-month pre-index period: descriptive analysis.

Figure 3. Direct and indirect work-loss costs during the 12-month post-index period: descriptive analysis.

Table 1. Sample selection criteria.

Steps	Selection criteria	Number of persons
1	Number of beneficiaries	2,853,441
2	At least one prescription of selective serotonin reuptake inhibitors, serotonin/norepinephrine reuptake inhibitors or bupropion	339,024
3	At least one diagnosis of major depressive disorder [ICD-9 code: 296.2x, 296.3x]	60,990
4	Employees only	25,538
5	Age 18–64, and index date on or before December 31, 2003 (to provide a 12-month study period ending December 31, 2004)	22,798
6	Continuous eligibility for 6 months prior to the index date and 12 months thereafter	6,106
7	Patients excluded if had any antidepressants within 6 months prior to the index date	5,489
8	Patients excluded if did not have at least one day of index drug monotherapy on index date	5,225

Table 2. Distribution of patients by treatment pattern and stability group.

Initial period	Subsequent period					Total for initial period
	Switchers	Add-on	Discontinuers	Maintainers	No treatment
Switchers	109	76	79	176	93	533
	2.1%	1.5%	1.5%	3.4%	1.8%	10.2%
Add-on	44	143	35	78	8	308
	0.8%	2.7%	0.7%	1.5%	0.2%	5.9%
Discontinuers	50	38	677	133	787	1,685
	1.0%	0.7%	13.0%	2.6%	15.1%	32.3%
Maintainers	193	116	762	1,619	9	2,699
	3.7%	2.2%	14.6%	31.0%	0.2%	51.7%
Total for subsequent period	396	373	1,553	2,006	897	5,225
	7.6%	7.1%	29.7%	38.4%	17.2%	100.0%
	Non-Stable		Intermediate		Stable
	n = 769		n = 1,279		n = 3,177
	14.7%		24.5%		60.8%

Table 3. Study sample characteristics and medical profiles in the 6-month pre-index period by treatment stability group.

Study sample	Baseline
	Stable	Intermediate	Non-stable	p-value
N=5,225 (% total)	3,177 (60.8%)	1,279 (24.5%)	769 (14.7%)
Demographic factors
Age, mean (sd)	41 (9)	41 (9)	41 (9)	0.057
Gender, % male	39.4	36.3	38.6	0.151
Medical profiles (rate %)
Non-depression mental comorbidities diagnosed	9.3	10.2	11.4	0.191
Any type of anxiety	5.3	5.6	6.8	0.296
Generalised anxiety disorder	2.5	2.7	2.2	0.820
Panic disorder	1.4	1.6	2.2	0.220
Post-traumatic stress disorder	0.6	0.8	1.2	0.237
Phobia	0.7	0.9	1.2	0.335
Obsessive-compulsive disorder	0.7	0.5	0.7	0.751
Sleeping disorder	2.6	2.7	2.7	0.969
Bipolar	1.4	1.9	2.2	0.192
Other psychotic disorders	0.4	0.5	0.5	0.902
Eating disorder	0.3	0.1	0.1	0.349
Schizophrenia	0.0	0.2	0.1	0.333
Physical comorbidities diagnosed	11.4	10.6	10.0	0.485
Deyo-Charlson Comorbidity Index, mean (SD)	0.25 (0.97)	0.20 (0.79)	0.20 (0.85)	0.453
Substance abuse diagnosed	2.5	3.6	3.5	0.077
Drug abuse/dependence	1.4	2.1	2.1	0.173
Alcohol abuse/dependence	1.2	1.6	1.7	0.371
Injuries/accidents	21.1	22.0	19.0	0.271
Urgent care	13.5	15.5	15.5	0.129
Hospitalisations	8.3	10.0	10.7	0.046
Emergency room visits (for any reason)	6.5	6.7	6.6	0.977

SD, standard deviation.

Table 4. Multivariate analysis of direct and indirect work-loss costs during the 12-month post-index period.

Work-loss costs	Stable		Intermediate		Non-stable		Stable vs. intermediate	Stable vs. non-stable	Intermediate vs. non-stable	Stable vs. intermediate	Stable vs. non-stable	Intermediate vs. non-stable
	(1)		(2)		(3)		(2) – (1)	(3) – (1)	(3) – (2)	(2)–(1)	(3)–(1)	(3)–(2)
	Mean	SE	Mean	SE	Mean	SE		Cost savings		p-value	p-value	p-value
Total direct costs, $	6,215	262	7,317	364	9,948	523	1,102	3,732	2,631	<0.0001	<0.0001	<0.0001
Drug costs, $	1,618	56	1,603	76	2,663	116	–15	1,044	1,059	0.001	<0.0001	<0.0001
Medical costs, $	4,618	240	5,690	332	7,277	474	1,073	2,660	1,587	<0.0001	<0.0001	<0.0001
Total work loss costs, $	2,221	65	2,829	146	4,125	226	608	1,904	1,296	<0.0001	<0.0001	<0.0001
Absenteeism costs, $	1,414	35	1,569	54	2,056	97	155	643	487	<0.0001	<0.0001	<0.0001
Disability costs, $	803	56	1,278	128	2,060	209	474	1,256	782	<0.0001	<0.0001	<0.0001
Total direct and work-loss costs	8,411	289	10,078	451	14,128	640				<0.0001	<0.0001	<0.0001

SE, standard error.

Notes

1. Sum of stable, intermediate, and non-stable sample size is 5,225.

2. SE and p-value based on a bootstrap method.

3. Direct costs are the sum of medical costs and drug costs.

4. The variables included in the regression are: age, gender, index year, index therapy, baseline mental comorbidities, the baseline Deyo-Charlson Comorbidity Index, and baseline severity markers (6-month pre-index date rate of hospitalisation, emergency room visits, injuries/accidents, drug abuse, and alcohol abuse).

5. A bootstrap method shows statistical significance in comparisons of stable, intermediate, and non-stable for all costs. The statistical tests assessed total direct and work loss costs, as well as individual cost components, in the 12-month post-index period.

Acknowledgements

Declaration on interest: This research was supported by sanofi-aventis. Analysis Group has received research support from sanofi-aventis and Caroline Amand is an employee of sanofi-aventis.

This study was presented at the 2007 American Psychiatric Association Annual Meeting, San Diego, CA, May 19–24, 2007.

Notes