This document discusses a study comparing psychiatric re-hospitalization rates across countries using routine health care data. The study, called CEPHOS-LINK, analyzed data from six European countries to identify predictors of re-hospitalization and compare rates. Ensuring the data was interoperable between countries with different health systems and data collection practices was challenging. The study found re-hospitalization rates varied between countries and psychosis was a consistent predictor of higher rates. Continuity of outpatient care after discharge also impacted rates. However, differences in health systems, data quality, and definitions of variables like hospital discharge limited full comparability.

1. Comparing psychiatric re-hospitalization rates
across countries by using
routine health care data:
possibilities and limitations
Heinz Katschnig, MD
Professor of Psychiatry
IMEHPS.research, Vienna, Austria
heinz.katschnig@imehps.at
www.imehps.at

2. Contents
• Why compare psychiatric rehospitalisation
rates across countries?
• How was it done?
• What are the findings?
• Conclusions

3. Rehospitalisation rates to the same hospital are used in
the US as performance indicator of hospital care and
hospitals with more than average rates are penalized

4. Problems with this performance indicator
Hospital Readmission Rates
• Only readmission to the same hospital =
20% less than total readmission rates
• Case mix different in different hospitals
• Fewer deaths in hospitals > more readmissions
• Communtiy care matters, not only hospital care
Nevertheless ……. The OECD publishes such rates
suggesting that they are an indicator of comparing
quality of care in differernt countries

5. Schizophrenia 30 days readmission rates
to the same hospital 2009, OECD 2011 > Implicit rating!

6. Problems with these Hospital Readmission Rates
In times of shift to community psychiatry >
Are high psychiatric re-hospitalisation rates a sign of
quality of mental health care?
For hospital care? For community care?
Can we compare countries and implicitly assess the
performance of the mental health care system of
different countries as the OECD does?

7. • Are such differences in re-hospitalisation rates real?
• Or do they reflect methodological differences?
• Can predictors be identified in a consistent way?
• Problem of systematic reviews and meta-analyses:
each study has different design
• Literature reviews have found few consistent results

8. Contents
• Why compare psychiatric rehospitalisation
rates across countries?
• How was it done?
• What are the findings?
• Conclusions

9. This project has received funding from the European Union’s Seventh Framework Programme for research,
technological development and demonstration under grant agreement no 603264
Comparative Effectiveness research on Psychiatric
HOSpitalisation
by record LINKage of large administrative data sets
coordinated by THL
36 month: 4/2014 – 3/2017
Austria, Finland, Italy, Norway, Romania and Slovenia
EU- FP7 - HEALTH F3-2014 - 603 264

10. Participants
3 Mandatory Health Insurance Countries
Austria (IMEHPS, dwh),
Romania (SNSPMPDS),
Slovenia (Acadamy of Science)
3 National Health Service Countries
Finland (THL, coordinator)
Italy (University of Verona),
Norway (SINTEF)

12. About CEPHOS-LINK 1: General
• Comparing re-hospitalisation rates of patients with a main
psychiatric diagnosis ICD-10 F2-FG after discharges from a
“psychiatric inpatient service”
• Identifying predictors for re-hospitalisation rates
• Using routine data from large national electronic administrative
registries with personally linked records
• Performing analyses in each country separately and with a pooled
data set
• Multiple logistic and Cox regression analyses in a “retrospective
cohort study design”
• Simulation modelling (“what would happen, if ….. demographics
etc. change”)

13. About CEPHOS-LINK 2:
Advantages if compared to tradional studies
• Including all hospitals in a country – unselected total
population –
in CEPHOS-LINK a total of 225.600 patients
• Reduce „methodological noise“ by ascertaining
interoperability of used databases
• Identifying comparable study cohorts
• Accounting for case mix and regional factors
• Assessing continuity of care in the community

14. About CEPHOS-LINK 3:
Disadvantages compared to traditional studies
• In routine databases only limited number of variables
available, even more reduced by need to identify
common denominator across countries (future: data
mining of „Electronic Health Records“?)
• Limited granularity of available variables
• Ethical and legal problems have to be cleared
• Ascertaining interoperability of databases took half of
the project time

17. Not to forget in general
• Data pooling comes for clinical trials – but problem: different sizes of study
population
• International health care statistics = counting events
• Advantage of routine health care data
large unselected whole country populations
• Linking person level data > pathways – re-hospitalisation as quality
indicator, to same hospital, to all hospitals = plus 20%, check aftercare
• Ethical problems, anonymity
• Health care system – TAX vs. Insurance – egebins pooling zeigen
• Pooling > quality of data improved
• Prospective cohort study vs retrospective cohort study vs EHR – see
EUROCOHORTS

18. Nor to forget hospital
• In and out of a psychiatric hospital was clear – foto
Today many types of hospitals – what does it mean – transfer – häusln von
christa permeability between different departements, day care,
comorbidity – show map
• International health care statistics = couting events as opposed to
pathways – re-hospitalisation
heavy users, gate keeping

19. Outline
1. The problem of comparability of routine health care data across
different countries
2. About the CEPHOS-LINK study
3. Ensuring interoperability of data and examples of reporting
mechanisms’ influence on health care data
4. Examples of the influence of health systems and provider payment
on the case mix of study cohorts for rehospitalisation studies
5. Discussion, lessons learned and outlook

20. Ensuring interoperability in order not to
compare “apples with oranges”
• Harmonising terminology, concepts and definitions and check
how they are operationalised in the actual data
• Obtaining background information (e.g. purpose of the database)
• Understanding the health care system - differences in service
organisation and provision in the different countries – e.g.
financing mechanisms
• Data/variable quality check (e.g. obtaining frequencies,
validating/comparing with national statistics, exploring coding
practices)
• What is in, what is out - inclusion and exclusion of data/variables
in the database
– inclusion / exclusion of patient groups /populations
– inclusion / exclusion of service providers
– inclusion / exclusion of services provided / utilisation

21. Outline
1. The problem of comparability of routine health care data across
different countries
2. About the CEPHOS-LINK study
3. Ensuring interoperability of data and examples of reporting
mechanisms’ influence on health care data
4. Examples of the influence of health systems and provider payment
on the case mix of study cohorts for rehospitalisation studies
5. Discussion, lessons learned and outlook

22. The problems with comparability of routine
health care data from different countries
While the big advantage is large unselected patient populations:
• Data is usually not collected for research purposes
• Data has already been collected – analyses depend on variables
included
• Differences in inclusion of service types, populations, utilisation
records
• Quality of data varies, e.g., due to
– Differences in coding routines
– Differences in health care organisation (e.g. payment mechanisms)
– Data flow (depending on legal, organisational, administrative issues of a
country)
– Differences in variables included/excluded
– Differences in data granularity

23. Outline
1. The problem of comparability of routine health care data across
different countries
2. About the CEPHOS-LINK study
3. Ensuring interoperability of data and examples of reporting
mechanisms’ influence on health care data
4. Examples of the influence of health systems and provider payment
on the case mix of study cohorts for rehospitalisation studies
5. Discussion, lessons learned and outlook

24. Outline
1. The problem of comparability of routine health care data across
different countries
2. About the CEPHOS-LINK study
3. Ensuring interoperability of data and examples of reporting
mechanisms’ influence on health care data
4. Examples of the influence of health systems and provider payment
on the case mix of study cohorts for rehospitalisation studies
5. Discussion, lessons learned and outlook

25. Challenges with different concepts,
terminologies and reliability of variables
• Different meanings of the concepts discharge and “transfer”
• Difficulties in identifying inter- and intra-hospital transfers in
the databases
• Variables and codes identifying a discharge and admission
were not reliable (e.g. different coding cultures,
inconsistencies in coding)
• Difficulties in identifying “hospital” in a comparative way
>>>consequences for calculating length of stay in a
comparative way

26. Hospital Separation Types
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Austria
Finland
Italy
Norway
Romania
Slovenia
Discharges Deaths Transfers
TRANSFERS?

27. Examples of problems
of how to define a hospital discharge

28. Example 1: Intra-hospital transfer

29. Example 2: inter-hospital transfer

30. Example 3: intra-organisational transfer

31. Separation codes (end of a hospital stay)
Finland
3 = Dead
1 = Institutions
11 = Transfer to hospital
12 = Transfer to primary care ward in community health centre
13 = Transfer to nursing home
14 = Transfer to institution for people with learning disability
15 = Transfer to institution for people with substance abuse
16 = Transfer to institution for rehabilitation
18 = Transfer to other institutions
2 = Home and home-based care
21 = Transfer to care at home/supported housing without 24h supervision
22 = Transfer to home without repeated care
23 = Transfer to supported housing (24h support) for old people
24 = Transfer to supported housing for people with learning disabilities

32. Separation codes (end of a hospital stay)
Romania
1 = discharged
2 = discharged on request
3 = Transfer to another hospital
4 = died

33. Finnland
In the HILMO database a large proportion of patients
had a transfer code to a different hospital (11), much
larger than in other countries
However, when using record linkage a large
proportion did not show up in a different hospital on
the same day
Rules were changed for defining the CEPHOS-LINK
study cohort – separation codes were not used, but
record linkage was used to exclude patients for follow-
up (in addiotion to excluding patients who had died

34. Types of psychiatric inpatient service
33
49
8
24
27
83
67
51
89
15
73
17
0 0
3
61
0 0
0
10
20
30
40
50
60
70
80
90
100
Austria
N=27
Finland
N=49
Italy
N=351
Norway
N=151
Romania
N=73
Slovenia
N=6
Percentage of 3 main types of psychiatric inpatient services
in the six CEPHOS-LINK countries
Stand alone psychiatric hospital and psychiatric departments not on the grounds of a general hospital
Psychiatric department on the grounds of a general hospital
Psychiatric centre / community mental health centre
Departments
in General
Hospitals
District
psychiatric
centres
Standalone
psychiatric
hospitals

35. „Fake“ Psychiatric Department in a
General Hospital
Muurolan
Sairaala

36. Mapping psychiatric inpatient services

37. Outline
1. The problem of comparability of routine health care data across
different countries
2. About the CEPHOS-LINK study
3. Ensuring interoperability of data and examples of reporting
mechanisms’ influence on health care data
4. Examples of the influence of health systems and provider payment
on the case mix of study cohorts for rehospitalisation studies
5. Discussion, lessons learned and outlook

38. Contents
• Why compare psychiatric rehospitalisation
rates across countries?
• How was it done?
• What are the findings?
• Conclusions

39. 30 and 365 days
psychiatric re-hospitalisation rates
16%
10% 10%
15%
8% 9%
40% 40%
36%
48% 46%
34%
0%
10%
20%
30%
40%
50%
60%
Österreich
N=21.839
Finnland
N=16.814
Italien
N=63.419
Norwegen
N=17.158
Rumänien
N=101.834
Slovenien
N=4.536
30 Tage 365 Tage

40. Psychiatric re-hospitalisation

41. 8 predictors
Patient-related predictors
• Female Gender
• AgeOld > Median
• LOS > Median
• Psychosis (ICD-10 F2, F30, F31)
• Physical comorbidity
Contextual predictors NUTS 3 level
• Urbanicity
• GDP
Continuity of care
• Psychiatric outpatient contact post-discharge

42. Logistic regression 30 and 365 days
psychiatric re-hospitalisations
Austria, Italy,
Romania, Slovenia
Finland, Norway
30d 365d 30d 365d
Gender 0 0 0 0
AGEOLD - - - - - -
LOSLONG_ALL 0 0 - - -
PSYCHOSIS + ++ + ++
PHY_COMORB - 0 0 0

43. Odds ratios for psychosis 30 and 365 days
re-hospitalisation
1,27
1,09
2,27
0,93
1,75
1,24
1,54
1,50
0
0,5
1
1,5
2
2,5
AT IT RO SI
30d 365d

44. First psychiatric outpatient contact
in first three weeks after discharge

45. First psychiatric re-hospitalisation
in first three weeks after discharge

46. Heterogenity of diagnostic mix
Provider payment influence on diagnosis coding
in the Veneto Region? (1)
31%
3%
1%
43%
12%
11%
14%
1%
13%
18%
16%
39%
0% 5% 10% 15% 20% 25% 30% 35% 40% 45%
F6 Personality disorder
F5 Psychsomatic
F4 Anxiety etc.
F32-F39 Depression
F30-F31 Bipolar
F2 Schizophrenia
Percentage of patients by F2-F6 main diagnosis in Veneto by public and
private providers
PUBLIC
N=5.062
PRIVATE
N=2.177

47. Heterogenity of diagnostic mix
Provider payment influence on diagnosis coding
in the Veneto Region? (2)
Public providers - more psychotic patients
Payment mechanism:
• global budgets
Private providers - fewer psychotic patients
Payment mechanism:
• Per diems
• Main diagnosis determines how many days of care are
maximally paid, e.g. affective disorders (F3): 30 days,
personality disorders (F6): 90 days
>>>>Incentive to admit “easy” patients with longer length of stay
(cream skimming)

48. Psychiatric re-hospitalisation in Veneto
by public and private providers

49. Psychiatric re-hospitalisation in Veneto
by public and private providers and total

50. Provider payment influence on length of stay?
15
11
32
0
5
10
15
20
25
30
35
AT
N=21.839
IT
N=63.419
SI
N=4.536
Median length of stay for the index episode including a spell on a
psychiatric hospital bed

51. Provider payment influence on length of stay?
(1)
Austria DRG (LKF-System)
• Fixed pool of funds for in-patient care
in a federal state per year
• Ex-post determination of the
monetary value for one point
• Distribution among hospitals
according to points accrued
• Incentives for hospital owners to
accrue more points (costlier
diagnoses, more episodes, shorter
LoS)
15
11
32
0
5
10
15
20
25
30
35
AT
N=21.839
IT
N=63.419
SI
N=4.536
Median length of stay for the
index episode including a
spell on a psychiatric hospital
bed

52. Provider payment influence on length of stay?
(2a)
Italy
Public general hospitals
• Global budgets
• No incentives concerning the number
of episodes
Private psychiatric hospitals
• Per diems
• Main diagnosis determines how many
days of care are maximally paid, e.g.
affective disorders (F3): 30 days,
personality disorder (F6): 90 days
15
11
32
0
5
10
15
20
25
30
35
AT
N=21.839
IT
N=63.419
SI
N=4.536
Median length of stay for the
index episode including a
spell on a psychiatric hospital
bed

53. Provider payment influence on length of stay?
(2b)
10
15
11
29
0
5
10
15
20
25
30
35
IT public IT private VEN public VEN private
Median length of stay for the index episode including a spell on a
psychiatric hospital bed in Italy and Veneto by public and private
providers

54. Provider payment influence on length of stay?
(3)
Slovenia: Flat-rate per episode
• Fixed number of episodes per year set
by SHI, historically determined
• Hospitals are not reimbursed for any
episodes beyond the contracted
number and receive less money if the
number of episodes is lower than
agreed
• Incentives for hospital owners to reach
exactly the contracted number of
episodes (not more, not fewer)
15
11
32
0
5
10
15
20
25
30
35
AT
N=21.839
IT
N=63.419
SI
N=4.536
Median length of stay for the
index episode including a
spell on a psychiatric hospital
bed

55. 37,62
8,78
14,74
5,23
44,89
5,36
0
5
10
15
20
25
30
35
40
45
50
AT (2006)
N=21.839
FI (2012)
N=16.814
IT (2012)
N=63.419
NO (2012)
N=17.158
RO (2012)
N=93.450
SI (2013)
N=4.536
Percentage
Patients of the study cohort with physical comorbidity
Coded additonal physical diagnoses
Differences in coding additional physical diagnoses

56. Romanian case mix effect responsible for psychiatric
re-hospitalisation rate curve in months 11-12 (1)

57. Romanian case mix effect responsible for psychiatric
re-hospitalisation rate curve in months 11-12 (2)
Possible explanation for increase of re-
hospitalisation rates in Romania between 330
and 365 days:
Official requirements to have at least one
inpatient admission in psychiatry in order to be
able to re-apply for a disability pension due to a
psychiatric diagnosis

58. Outline
1. The problem of comparability of routine health care data across
different countries
2. About the CEPHOS-LINK study
3. Ensuring interoperability of data and examples of reporting
mechanisms’ influence on health care data
4. Examples of the influence of health systems and provider payment
on the case mix of study cohorts for rehospitalisation studies
5. Discussion, lessons learned and outlook

59. Contents
• Why compare psychiatric rehospitalisation
rates across countries?
• How was it done?
• What are the findings?
• Conclusions

60. Discussion (1)
• The data used to construct the performance
indicator “psychiatric re-hospitalisation” is
depending on multiple factors, which again are
interacting with the way health care systems are
organised, providers paid, and data recorded and
reported to national databases.

61. Discussion (2)
• Implications for Health Care Provision and Use: In
cross country comparison the knowledge and
awareness of which data stand for the quality of
health care and which data don’t, should help to put
results on health care provision and use into
perspective.
• Implications for Health Policies: The influence of
health care organisation, provider payment and data
reporting mechanisms on health care service
utilization data needs to be considered when re-
hospitalisation rates are intended to be used.

62. Lessons learned
Ensuring interoperability is a major issue
when integrating registry based patient cohorts
across different countries
– Involves numerous steps (e.g. harmonising terminology, pilot
analyses) and investigative approaches in finding out what data
stands for (e.g. on-site visits with data-owners, health care
providers) and is most time-consuming
– A good knowledge of the health care system, its organisation and
payment mechanisms are essential in order to check plausibility
of data, results and interpretation of results
– Separate analyses between region, different providers,
etc. may help to explore the influence of health system factors
– Plausibility checks with existing national statistical data

63. Lessons learned
Understanding data and its limitations – what is in it,
what is left out, how comparable is data. Differences in
the databases may concern
– the inclusion and exclusion of patient groups/populations
– the inclusion and exclusion of service providers (e.g. “Wahlärzte” in
Austria)
– the inclusion and exclusion of service utilisation records
(utilisation of different types of care e.g. rehabilitation)

64. Outlook
With the rise of big data stored in large
electronic databases the opportunity of sharing
and pooling health care utilisation data and
using it for cross-country comparison has
increased and will gain even more importance in
the future.
> Ensuring interoperability of data is essential
and such an endeavour.

65. AHA - effect
• By using linked routine data one can get get a
look into details of health care system and its
functioning which are usually overlooked
E.g. Private psychiatric hospitals in Italy
Fake psychiatric departments in
General hospitals
Different case mix: Depression or
schizophrenia dominating
• New questions can be asked – what is truth?

66. Thank you
for
your attention!
www.cephos-link.org
christa.strassmayr@imehps.at
IMEHPS.research, Vienna, Austria
www.imehps.at