MORTALITY IN PERSONS WITH MENTAL DISORDERS IS SUBSTANTIALLY OVERESTIMATED USING INPATIENT PSYCHIATRIC DIAGNOSES

Abstract

Mental disorders are associated with premature mortality, and the magnitudes of risk have commonly been estimated using hospital data. However, psychiatric patients who are hospitalized have more severe illness and do not adequately represent mental disorders in the general population. We conducted a national cohort study using outpatient and inpatient diagnoses for the entire Swedish adult population (N=7,253,516) to examine the extent to which mortality risks are overestimated using inpatient diagnoses only. Outcomes were all-cause and suicide mortality during 8 years of follow-up (2001–2008). There were 377,339 (5.2%) persons with any inpatient psychiatric diagnosis, vs. 680,596 (9.4%) with any inpatient or outpatient diagnosis, hence 44.6% of diagnoses were missed using inpatient data only. When including and accounting for prevalent psychiatric cases, all-cause mortality risk among persons with any mental disorder was overestimated by 15.3% using only inpatient diagnoses (adjusted hazard ratio [aHR], 5.89; 95% CI, 5.85–5.92) vs. both inpatient and outpatient diagnoses (aHR, 5.11; 95% CI, 5.08–5.14). Suicide risk was overestimated by 18.5% (aHRs, 23.91 vs. 20.18), but this varied widely by specific disorders, from 4.4% for substance use to 49.1% for anxiety disorders. The sole use of inpatient diagnoses resulted in even greater overestimation of all-cause or suicide mortality risks when prevalent cases were unidentified (~20–30%) or excluded (~25–40%). However, different methods for handling prevalent cases resulted in only modest variation in risk estimates when using both inpatient and outpatient diagnoses. These findings have important implications for the interpretation of hospital-based studies and the design of future studies.

INTRODUCTION

Research from the past 20 years has shown that mental disorders are associated with increased all-cause(Harris & Barraclough, 1998) and suicide(Harris & Barraclough, 1997) mortality. Most large, highly cited studies have ascertained mental disorders using hospital discharge data, which are more commonly available than outpatient data.(Hamer, Stamatakis, & Steptoe, 2008; Hiroeh, Appleby, Mortensen, & Dunn, 2001; Hiroeh et al., 2008; Hoang, Stewart, & Goldacre, 2011; Laursen, Munk-Olsen, Nordentoft, & Mortensen, 2007; Politi et al., 2002; Qin, Agerbo, & Mortensen, 2003; Wahlbeck, Westman, Nordentoft, Gissler, & Laursen, 2011) However, persons with mental disorders who are hospitalized typically have more severe illness and do not adequately represent mental disorders in the general population. This is increasingly the case as more mental disorders are now treated in outpatient settings.(Narrow et al., 2000; Wahlbeck et al., 2011) Studies that ascertain mental disorders solely using inpatient diagnoses will miss many milder cases, likely resulting in overestimation of mortality risks for the general population,(Gale et al., 2012) although to what extent is unclear. Furthermore, studies have used different methods for handling prevalent psychiatric cases, which may either be identified and included, identified and excluded, or unidentified (counted as a case only if a new diagnosis). Although inclusion of prevalent cases may result in underestimation of mortality risks due to survivor bias, such bias may potentially be avoided if properly accounting for variable time at risk after diagnosis (left truncation),(Azzato et al., 2009; Brookmeyer, 2005) with the additional advantage of improved statistical power due to a larger sample size. The extent to which mortality risks are overestimated using only inpatient diagnoses, and the relative effects of different methods for handling prevalent cases, are unknown. Such information would have important implications for the interpretation of the large number of hospital-based studies, health policy planning based on such studies, and the design of future studies.

We used nationwide outpatient and inpatient data in Sweden to examine 1) the extent to which all-cause and suicide mortality risks in the general population are overestimated when mental disorders are ascertained using only inpatient compared with both inpatient and outpatient diagnoses, and 2) the extent to which risk estimates are influenced by different methods for handling prevalent psychiatric cases.

MATERIALS AND METHODS

Study Population and Outcomes

The study population consisted of all persons aged 17 years or older who were living in Sweden on January 1, 2001 (N=7,253,516). The outcomes were all-cause mortality and suicide mortality (International Classification of Diseases, version 10 [ICD-10] codes X60-X84) during 8 years of follow-up (January 1, 2001 through December 31, 2008). These outcomes were identified using the Swedish Death Registry, which includes ~99.5% of all deaths nationwide.(National Board of Health and Welfare, 2011) Suicide was examined because it is the most widely studied cause of death among psychiatric patients.(Harris et al., 1997) This study was approved by the Regional Ethical Review Board of Lund University in Sweden.

Mental Disorder Ascertainment

Mental disorders were identified by any primary or secondary diagnosis in the Swedish Outpatient Registry or the Swedish Hospital Registry during the same follow-up period (January 1, 2001 through December 31, 2008). The Swedish Outpatient Registry contains all primary and secondary outpatient diagnoses nationwide starting in 2001; and the Swedish Hospital Registry contains all primary and secondary inpatient psychiatric diagnoses starting in 1973 (non-psychiatric diagnoses in 1964) for six counties in southern Sweden, and with nationwide coverage since 1987. These registries were recently estimated to be >99% complete.(Ludvigsson et al., 2011) “Prevalent” mental disorders were identified by any primary or secondary diagnosis in the Swedish Hospital Registry from 1973 through 2000 (before nationwide outpatient data were available). Mental disorders were classified according to ICD-10 codes and examined in the following categories: any mental disorder (F00–F69), substance use disorders (F10–F19), schizophrenia (F20), bipolar disorder (F31), depression (F32–F33), anxiety disorders (F40–F41), and personality disorders (F60–F61). Other less common mental disorders (and specifically those with <100 suicides) were not examined separately. To evaluate excess mortality risks associated with having more than one psychiatric condition, we also examined “any second mental disorder” among those specified above, while adjusting for any first mental disorder.

Statistical Analysis

Cox proportional hazards regression was used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the association between mental disorders and either all-cause or suicide mortality. Each model was performed alternatively using only inpatient or both inpatient and outpatient diagnoses during the follow-up period (2001–2008). Prevalent psychiatric cases (diagnosed in 1973–2000) were alternatively: 1) identified and included, accounting for left truncation (variable time at risk after diagnosis) in the Cox model(Brookmeyer, 2005); 2) unidentified, i.e., diagnoses prior to 2001 were disregarded, so that persons were counted as cases only if they received a diagnosis during 2001–2008; or 3) identified and excluded from the model. Mental disorders were modeled as time-dependent variables in each model, so that persons contributed to “non-exposed” person-time before and “exposed” person-time after the earliest diagnosis of the respective mental disorder. Persons with multiple psychiatric diagnoses were included in multiple categories so that they contributed to the risk estimate for each respective disorder. All hazard ratios were adjusted for age (modeled as a time-dependent continuous variable) and sex. Analyses of “any second mental disorder” were also adjusted for any first mental disorder by modeling these simultaneously as separate time-dependent variables, in order to estimate the excess mortality risk associated with a second disorder above and beyond that of a first disorder. Individuals were censored at the time of emigration, as determined by the absence of a Swedish residential address in census data (n=202,769; 2.8%); and, in analyses of suicide, at the time of death from any other cause (n=724,350, 10.0%). The proportional hazards assumption was evaluated by graphical assessment of log-log plots(Garrett, 1997) and was met in each of the models. All statistical tests were two-sided and used an α-level of 0.05. All analyses were conducted using Stata statistical software, version 11.2.(StataCorp, 2010)

RESULTS

In the entire population of 7,253,516 Swedish adults, 377,339 (5.2%) persons had an inpatient diagnosis of any mental disorder in 2001–2008, compared with 680,596 (9.4%) who had an inpatient or outpatient diagnosis of any mental disorder, hence 44.6% of diagnoses were missed using inpatient data only (Table 1). The proportion of diagnoses missed was highest for anxiety disorders (71.3%), and lowest for substance use disorders (27.3%) and schizophrenia (36.0%). Among cases identified using only inpatient diagnoses, 71,570 (19.0%) were prevalent (i.e., diagnosed in 1973–2000) which had an average follow-up of 9.6 years (maximum 21.4 years); whereas using both inpatient and outpatient diagnoses, there were 90,289 (13.3%) prevalent cases with an average follow-up of 9.8 years (maximum 22.0 years).

Table 1.

Number and prevalence of Swedish adults (N=7,253,516) with mental disorders ascertained alternatively using only inpatient or both inpatient and outpatient diagnoses (2001–2008).^a

Mental disorders (ICD-10 codes)	Inpatient diagnoses only n (%)	Inpatient and outpatient diagnoses n (%)	Cases missed using inpatient diagnoses only
Any mental disorder (F00–F69)	377,339 (5.2)	680,596 (9.4)	44.6%
Substance use disorders (F10–F19)	116,924 (1.6)	160,762 (2.2)	27.3%
Schizophrenia (F20)	16,620 (0.2)	25,970 (0.4)	36.0%
Bipolar disorder (F31)	14,855 (0.2)	30,895 (0.4)	51.9%
Depression (F32–F33)	95,492 (1.3)	214,830 (3.0)	55.5%
Anxiety disorders (F40–F41)	42,633 (0.6)	148,498 (2.0)	71.3%
Personality disorders (F60–F61)	12,952 (0.2)	32,330 (0.4)	59.9%
At least 2 mental disorders	44,824 (0.6)	115,206 (1.6)	61.1%

^aThese comparisons are based on psychiatric diagnoses from all inpatient and outpatient settings during 2001–2008, and do not include diagnoses prior to 2001 when nationwide outpatient data were unavailable.

All-Cause Mortality

When prevalent cases were included and left truncation accounted for, the adjusted hazard ratio (aHR) for association between any inpatient-diagnosed mental disorder and all-cause mortality was 5.89 (95% CI, 5.85–5.92), which was 15.3% higher and had a non-overlapping confidence interval compared with the corresponding estimate using both inpatient and outpatient diagnoses (aHR, 5.11; 95% CI, 5.08–5.14) (Table 2, first row). The sole use of inpatient diagnoses resulted in undercounting deaths among persons with mental disorders by >20,000 (12.8%). It also resulted in overestimation of all-cause mortality risk for each mental disorder examined separately, ranging from 6.7% for substance use disorders to 25.9% for anxiety disorders. Any second mental disorder was associated with more than a ~2.5-fold mortality risk beyond that associated with a first mental disorder, and this risk also was substantially overestimated (20.1%) when using only inpatient diagnoses. Specific mental disorders that occurred as a second psychiatric diagnosis were examined separately and had similar hazard ratios that were significantly elevated for each (data not shown). Risk estimates for any first mental disorder were negligibly affected by adjustment for any second disorder (i.e., were similar to those reported for “any mental disorder” in Table 2) and therefore are not shown separately.

Table 2.

Age- and sex-adjusted hazard ratios for associations between mental disorders and all-cause mortality (2001–2008).

Mental disorders (ICD-10 codes)	Inpatient diagnoses only				Inpatient and outpatient diagnoses				Overestimation of aHR using inpatient data only
	Deaths	aHR (95% CI)	Maximum differencea		Deaths	aHR (95% CI)	Maximum differencea
Any mental disorder (F00–F69)
Prevalent cases identified and includedb	139,236	5.89 (5.85, 5.92)		16.0%	159,705	5.11 (5.08, 5.14)		4.5%	15.3%
Prevalent cases unidentifiedc	139,236	6.64 (6.60, 6.68)			159,705	5.34 (5.31, 5.37)			24.3%
Prevalent cases identified and excludedd	114,751	6.83 (6.78, 6.88)			132,884	5.34 (5.30, 5.37)			27.9%
Substance use disorders (F10–F19)
Prevalent cases identified and includedb	24,895	4.81 (4.74, 4.87)		19.8%	28,464	4.51 (4.45, 4.57)		14.0%	6.7%
Prevalent cases unidentifiedc	24,895	5.76 (5.68, 5.84)			28,464	5.14 (5.07, 5.20)			12.1%
Prevalent cases identified and excludedd	17,297	5.44 (5.35, 5.53)			20,302	4.76 (4.69, 4.84)			14.3%
Schizophrenia (F20)
Prevalent cases identified and includedb	3,742	4.31 (4.17, 4.46)		23.9%	4,498	3.78 (3.66, 3.90)		11.6%	14.0%
Prevalent cases unidentifiedc	3,742	5.34 (5.16, 5.53)			4,498	4.22 (4.08, 4.35)			26.5%
Prevalent cases identified and excludedd	2,369	5.34 (5.11, 5.59)			2,987	4.01 (3.85, 4.17)			33.2%
Bipolar disorder (F31)
Prevalent cases identified and includedb	2,650	3.11 (2.99, 3.24)		17.4%	3,485	2.53 (2.44, 2.63)		6.8%	22.9%
Prevalent cases unidentifiedc	2,650	3.65 (3.50, 3.81)			3,485	2.68 (2.58, 2.78)			36.2%
Prevalent cases identified and excludedd	1,828	3.63 (3.45, 3.82)			2,545	2.51 (2.40, 2.62)			44.6%
Depression (F32–F33)
Prevalent cases identified and includedb	25,981	3.65 (3.61, 3.69)		15.9%	33,207	3.20 (3.16, 3.23)		4.4%	14.1%
Prevalent cases unidentifiedc	25,981	4.12 (4.06, 4.17)			33,207	3.31 (3.27, 3.35)			24.5%
Prevalent cases identified and excludedd	23,072	4.23 (4.17, 4.29)			29,772	3.34 (3.30, 3.39)			26.6%
Anxiety disorders (F40–F41)
Prevalent cases identified and includedb	9,339	3.70 (3.63, 3.78)		15.7%	14,811	2.94 (2.89, 2.99)		3.1%	25.9%
Prevalent cases unidentifiedc	9,339	4.28 (4.18, 4.38)			14,811	3.03 (2.98, 3.09)			41.3%
Prevalent cases identified and excludedd	8,492	4.28 (4.18, 4.38)			13,729	2.98 (2.93, 3.04)			43.6%
Personality disorders (F60–F61)
Prevalent cases identified and includedb	1,514	4.10 (3.91, 4.31)		26.8%	2,238	3.33 (3.19, 3.48)		6.9%	23.1%
Prevalent cases unidentifiedc	1,514	5.12 (4.84, 5.42)			2,238	3.56 (3.40, 3.73)			43.8%
Prevalent cases identified and excludedd	1,100	5.20 (4.86, 5.55)			1,755	3.36 (3.19, 3.54)			54.8%
Any second mental disordere
Prevalent cases identified and includedb	7,697	3.17 (3.09, 3.25)		17.2%	11,982	2.64 (2.58, 2.70)		6.5%	20.1%
Prevalent cases unidentifiedc	7,697	3.71 (3.62, 3.80)			11,982	2.81 (2.75, 2.87)			32.0%
Prevalent cases identified and excludedd	6,483	3.58 (3.47, 3.70)			10,500	2.58 (2.52, 2.65)			38.8%

^aMaximum difference in aHRs among three different methods for handling prevalent psychiatric cases.

^bPsychiatric cases diagnosed before 2001 were included and left truncation (variable time at risk after diagnosis) accounted for in the Cox model.

^cPsychiatric diagnoses prior to 2001 were disregarded, so that persons were counted as cases only if they received a diagnosis during 2001–2008.

^dPersons with the respective psychiatric diagnoses prior to 2001 were excluded from the model.

^eRisk estimates for any second mental disorder were adjusted for any first mental disorder (modeled simultaneously as separate time-dependent variables), to estimate excess mortality risk associated with a second disorder above and beyond that of a first disorder.

Abbreviations: aHR = adjusted hazard ratio; CI = confidence interval.

We further examined the extent to which all-cause mortality risk was overestimated by using only inpatient diagnoses, while comparing different methods for handling prevalent cases. For each category of mental disorders, overestimation resulting from the sole use of inpatient data was consistently lowest when prevalent cases were included and left truncation accounted for (~5–30% range), intermediate when prevalent cases were unidentified (~10–45%), and greatest when prevalent cases were excluded (~10–55%) (see Table 2, last column).

In addition, we examined the extent to which risk estimates varied among these different methods for handling prevalent cases, while alternatively using only inpatient or both inpatient and outpatient diagnoses. For each category of mental disorders, these different methods resulted in substantial variation of risk estimates when using only inpatient diagnoses (maximum differences ~15–30%), but only modest variation when using both inpatient and outpatient diagnoses (consistently <15%, and <7% for most mental disorders) (see Table 2, “Maximum difference” columns).

Suicide Mortality

Among 8,821 persons who died from suicide, 3,179 (36.0%) had any inpatient psychiatric diagnosis during 2001–2008, compared with 3,875 (43.9%) who had any inpatient or outpatient psychiatric diagnosis (Table 3). Suicides among persons with mental disorders were therefore undercounted by nearly 700 (18.0%) when using inpatient data only.

Table 3.

Age- and sex-adjusted hazard ratios for associations between mental disorders and suicide mortality (2001–2008).

Mental disorders (ICD-10 codes)	Inpatient diagnoses only				Inpatient and outpatient diagnoses				Overestimation of aHR using inpatient data only
	Suicides	aHR (95% CI)	Maximum differencea		Suicides	aHR (95% CI)	Maximum differencea
Any mental disorder (F00–F69)
Prevalent cases identified and includedb	3,179	23.91 (22.84, 25.03)		14.9%	3,875	20.18 (19.28, 21.11)		11.8%	18.5%
Prevalent cases unidentifiedc	3,179	27.48 (26.26, 28.76)			3,875	21.46 (20.53, 22.44)			28.1%
Prevalent cases identified and excludedd	2,114	26.85 (25.45, 28.32)			2,664	19.19 (18.24, 20.19)			39.9%
Substance use disorders (F10–F19)
Prevalent cases identified and includedb	1,325	17.99 (17.00, 19.05)		18.0%	1,564	17.23 (16.31, 18.20)		8.4%	4.4%
Prevalent cases unidentifiedc	1,325	20.26 (19.08, 21.51)			1,564	18.67 (17.65, 19.76)			8.5%
Prevalent cases identified and excludedd	914	21.23 (19.78, 22.78)			1,103	18.61 (17.43, 19.87)			14.1%
Schizophrenia (F20)
Prevalent cases identified and includedb	190	15.07 (13.13, 17.30)		21.0%	236	13.25 (11.66, 15.05)		8.7%	13.7%
Prevalent cases unidentifiedc	190	16.99 (14.71, 19.62)			236	12.63 (11.20, 14.25)			34.5%
Prevalent cases identified and excludedd	114	18.23 (15.15, 21.94)			150	13.73 (11.68, 16.14)			32.8%
Bipolar disorder (F31)
Prevalent cases identified and includedb	256	29.56 (26.24, 33.30)		23.2%	331	22.69 (20.38, 25.26)		1.9%	30.3%
Prevalent cases unidentifiedc	256	33.78 (29.82, 38.28)			331	23.02 (20.61, 25.71)			46.7%
Prevalent cases identified and excludedd	190	36.43 (31.53, 42.09)			260	23.12 (20.42, 26.18)			57.6%
Depression (F32–F33)
Prevalent cases identified and includedb	1,531	35.25 (33.39, 37.22)		20.1%	1,992	26.55 (25.24, 27.93)		4.7%	32.8%
Prevalent cases unidentifiedc	1,531	42.33 (39.98, 44.83)			1,992	27.81 (26.40, 29.30)			52.2%
Prevalent cases identified and excludedd	1,271	41.79 (39.28, 44.46)			1,684	26.57 (25.12, 28.10)			57.3%
Anxiety disorders (F40–F41)
Prevalent cases identified and includedb	492	23.38 (21.48, 25.44)		19.7%	882	15.68 (14.64, 16.80)		5.8%	49.1%
Prevalent cases unidentifiedc	492	27.98 (25.52, 30.68)			882	16.06 (14.95, 17.25)			74.2%
Prevalent cases identified and excludedd	419	27.09 (24.53, 29.92)			783	15.18 (14.08, 16.37)			78.5%
Personality disorders (F60–F61)
Prevalent cases identified and includedb	360	39.77 (35.98, 43.95)		39.6%	474	30.22 (27.61, 33.08)		8.4%	31.6%
Prevalent cases unidentifiedc	360	51.51 (46.29, 57.32)			474	32.75 (29.80, 35.99)			57.3%
Prevalent cases identified and excludedd	262	55.52 (49.02, 62.88)			367	31.61 (28.42, 35.17)			75.6%
Any second mental disordere
Prevalent cases identified and includedb	971	43.66 (42.10, 50.15)		22.4%	1,370	31.25 (29.39, 33.21)		8.7%	39.7%
Prevalent cases unidentifiedc	971	48.57 (45.34, 52.02)			1,370	33.97 (31.95, 36.10)			43.0%
Prevalent cases identified and excludedd	767	53.44 (48.97, 58.33)			1,130	32.29 (29.97, 34.79)			65.5%

^aMaximum difference in aHRs among three different methods for handling prevalent psychiatric cases.

^bPsychiatric cases diagnosed before 2001 were included and left truncation (variable time at risk after diagnosis) accounted for in the Cox model.

^cPsychiatric diagnoses prior to 2001 were disregarded, so that persons were counted as cases only if they received a diagnosis during 2001–2008.

^dPersons with the respective psychiatric diagnoses prior to 2001 were excluded from the model.

^eRisk estimates for any second mental disorder were adjusted for any first mental disorder (modeled simultaneously as separate time-dependent variables), to estimate excess mortality risk associated with a second disorder above and beyond that of a first disorder.

Abbreviations: aHR = adjusted hazard ratio; CI = confidence interval.

The sole use of inpatient diagnoses resulted in slightly greater overestimation of suicide risks than all-cause mortality, but this varied more widely by specific mental disorders. When prevalent cases were included and left truncation accounted for, the adjusted hazard ratio for association between any inpatient-diagnosed mental disorder and suicide mortality was 23.91 (95% CI, 22.84–25.03), which was 18.5% higher and had a non-overlapping confidence interval compared with using both inpatient and outpatient diagnoses (aHR, 20.18; 95% CI, 19.28–21.11) (Table 3, first row). Among specific disorders, overestimation of suicide risk ranged from 4.4% for substance use to 49.1% for anxiety disorders. Any second mental disorder was associated with more than a ~30-fold suicide risk beyond that associated with a first mental disorder, and this risk also was markedly overestimated (39.7%) when using only inpatient diagnoses. Specific mental disorders that occurred as a second psychiatric diagnosis were examined separately and the hazard ratios were markedly elevated for each (data not shown), and varied in a similar pattern as the overall hazard ratios for specific disorders in Table 3.

As with all-cause mortality, overestimation of suicide risk by using only inpatient diagnoses was consistently lowest when prevalent cases were included and left truncation accounted for (<5% to 50% range), intermediate when prevalent cases were unidentified (~5–75%), and greatest when prevalent cases were excluded (~10–80%) (Table 3, last column). In addition, different methods for handling prevalent cases resulted in substantial variation of risk estimates when using only inpatient diagnoses (maximum differences ~15–40%), but only modest variation when using both inpatient and outpatient diagnoses (consistently <12% for all mental disorders) (Table 3, “Maximum difference” columns).

DISCUSSION

In this large cohort study, we found that risk estimates for all-cause and suicide mortality in the general population were substantially overestimated when mental disorders were ascertained using only inpatient compared with both inpatient and outpatient diagnoses. The sole use of inpatient data missed 45% of all psychiatric diagnoses, as well as 13% of all deaths and 18% of suicides among persons with mental disorders. As a result, all-cause and suicide mortality risks among psychiatric patients were overestimated by ~15–20% when prevalent cases were included and left truncation accounted for, and ~20–40% when prevalent cases were unidentified or excluded. The overestimation of suicide risk varied widely by specific disorders, and was modest (<15%) for substance use disorders but much greater (~30–80%) for anxiety disorders, personality disorders, depression, and bipolar disorder. In addition, we found that different methods for handling prevalent cases resulted in substantial variation in risk estimates when using only inpatient diagnoses, but only modest variation when using both inpatient and outpatient diagnoses. When using only inpatient diagnoses, the inclusion of prevalent psychiatric cases with appropriate left truncation methods consistently reduced (but did not eliminate) overestimation of mortality risks, and has the additional advantage of increasing sample size and statistical power for detecting smaller effect sizes.

Premature mortality among persons with mental disorders is a major clinical and public health problem(Demyttenaere et al., 2004; Murray & Lopez, 1996; Thornicroft, 2011) and has received highly deserved research attention.(Gale et al., 2012; Hamer et al., 2008; Harris et al., 1997; 1998; Hiroeh et al., 2001; Hiroeh et al., 2008; Hoang et al., 2011; Laursen et al., 2007; Politi et al., 2002; Qin et al., 2003; Russ et al., 2012; Wahlbeck et al., 2011) Studies have consistently reported that virtually all mental disorders are associated with premature mortality from both unnatural(Harris et al., 1997; Hiroeh et al., 2001) and natural(Harris et al., 1998; Hiroeh et al., 2008) causes, with a shorter life expectancy by up to ~15–20 years compared with the general population.(Wahlbeck et al., 2011) The magnitudes of risk, however, have most commonly been estimated through large hospital-based studies.(Hamer et al., 2008; Hiroeh et al., 2001; Hiroeh et al., 2008; Hoang et al., 2011; Laursen et al., 2007; Politi et al., 2002; Qin et al., 2003; Wahlbeck et al., 2011) To our knowledge, the current study is the first to examine the extent to which such risk estimates may be systematically biased. The findings have implications for the interpretation of the large number of hospital-based studies, which may substantially overestimate mortality risks for the general population. We found that the extent of overestimation was greatest when prevalent psychiatric cases were excluded, intermediate when they were unidentified, and lowest (but remained substantial) when they were included while accounting for left truncation. Population-based studies that include outpatient as well as inpatient diagnoses are needed to enable more valid and generalizable risk estimates. Such estimates would enable a more accurate understanding of risk magnitudes and a more reliable basis for health policy planning. We recently conducted other studies using nationwide outpatient and inpatient data in Sweden to examine more comprehensively the risks of somatic comorbidities and mortality among persons with schizophrenia(Crump et al., 2012) or bipolar disorder (in press); and of suicide (in press), homicidal death(Crump et al., 2013), or accidental death (submitted) among persons with various mental disorders.

The current study has several limitations, including the inability to examine mental disorders that were undiagnosed, hence the reported prevalence of mental disorders still likely underestimated the true prevalence in the general population. However, because Sweden has universal health care access and diagnoses were obtained from all health care settings nationwide, ascertainment was markedly improved compared with most previous studies. Prevalent cases were likely underreported because they were based on inpatient data from an earlier period (1973–2000) when nationwide outpatient data were unavailable. Generalizability of our findings to countries with different health care systems and utilization rates is also uncertain. However, low and decreasing psychiatric hospitalization rates have been reported in other Western countries, as in Sweden.(Narrow et al., 2000; Priebe et al., 2005) A U.S. study reported that only ~17% of persons with severe mental illness and <1% of those with non-severe mental illness were hospitalized in one year of follow-up,(Narrow et al., 2000) suggesting that inpatient diagnoses would be similarly inadequate for obtaining valid risk estimates for the association between mental disorders and mortality in the general population.

Future large cohort studies of the health outcomes of mental illness should ascertain mental disorders from outpatient and inpatient sources whenever feasible. When outpatient data are unavailable, findings from the current study should be considered in the interpretation of risk estimates. Inclusion of prevalent cases using appropriate left truncation methods is recommended because it may reduce (though not eliminate) overestimation of risks, while also increasing statistical power to detect smaller effect sizes. Additional ascertainment of mental disorders through psychiatric interview or outpatient record review in a representative sample of the study population would also allow an assessment of the proportion of diagnoses missed by inpatient data and the effect on risk estimates.