Understanding and Using NAMCS and NHAMCS Data: A Hands-On Workshop

1
Understanding and Using NAMCS and NHAMCS Data
A Hands-On Workshop

• Part II-Advanced Programming Techniques
• Esther Hing

2
Overview

• Issues when trending NAMCS/NHAMCS data
• CHC data estimates
• Provider-level estimates
• Visit-level data aggregated to provider-level
  statistics
• Visits vs. patient estimates
• Summary

3
NAMCS/NHAMCS trend data
4
Survey content varies over time

• Variables routinely rotate on and off survey
• Be careful about trending diagnosis prior to 1979
  because of ICDA (based on ICD-8)
• Even after 1980- be careful about changes in
• ICD-9-CM
• Number of medications varies over years
• 1980-81 8 medications
• 1985, 1989-94 5 medications
• 1995-2002 6 medications
• 2003 and after--8 medications
• Medications coded according to MULTUM terminology
  in 2006, and according to the National Drug code
  Directory maintained by FDA in years before 2006
  are not comparable.
• Diagnostic therapeutic service checkboxes vary

5
PDF of Survey Content for the NAMCS and NHAMCS is
on webpagewww.cdc.gov/nchs/about/major/ahcd1.htm
6
Public Use Data File Documentation for each year
is another source

• Documentation includes
• A description of the survey
• Record format
• Marginal data (summaries)
• Various definitions
• Reason for Visit classification codes
• Medication generic names
• Therapeutic classes

7
Combining multiple years

• 2 year combinations are best for subpopulation
  analysis
• 3-4 year combinations for disease specific
  analysis
• Keep adding years until you have at least 30 raw
  cases in important cells
• RSE improves incrementally with the number of
  years combined

8
RSE improves incrementally with the number of
years combined

• RSE SE/x
• RSE for percent of visits by persons less than 21
  years of age with diabetes
• 1999 RSE .08/.18 .44 (44)
• 1998 1999 RSE .06/.18 .33 (33)
• 1998, 1999, 2000 RSE .05/.21 .24 (24)

9
Combining multiple settings

• NAMCS, hospital emergency department (ED), and
  outpatient department (OPD) data can be combined
  in one or multiple years
• NAMCS OPD variables virtually identical, many
  ED variables are same
• OPD and NAMCS should be combined to get estimates
  of ambulatory physician care especially for
  African-American, Medicaid or adolescent
  subpopulations
• Only NAMCS has physician specialty

10
Variance computations

• Survey design variables need to be identical
  across time and settings regardless of software
  used
• SUDAAN 3 4-stage design variables available for
  survey years 1993 through 2001
• Starting in 2002, 1-stage design variables were
  released with PUF files, permitting use of
  SUDAAN 1-stage WR variances, STATA, SASs Complex
  Survey procedures and SPSSs Complex Samples 12.0
  module

11
Design VariablesSurvey Years
2002
2001
1-Stage design variables 3- or 4-Stage design
variables
3- or 4-Stage design variables
2003
1-Stage design variables only
12
Code to create design variables survey years
2001 earlier
CPSUMPSUM CSTRATM STRATM IF CPSUM IN(1, 2,
3, 4) THEN DO CPSUM PROVIDER 100000 CSTRATM
(STRATM100000) (1000(MOD(YEAR,100)))
(SUBFILE100) PROSTRAT END ELSE CSTRATM
(STRATM100000)
13
2006 NAMCS Community Health Center data
14
NAMCS sample of Community Health Centers (CHCs)

• CHC physicians always included in NAMCS
• Typically small n of CHC physicians precluded
  presentation of estimates (unreliable)
• 2006 NAMCS included separate stratum of about 100
  CHCs
• Within CHCs, up to 3 physicians or mid-level
  providers (physician assistants or nurse
  practitioners) and their visits sampled

15
Comparison of primary care visits to community
health centers and physician offices

• 1/Difference between community health centers and
  physician offices is statistically significant
  (plt0.05).
• SOURCE Cherry DK, Hing E, Woodwell DA,
  Rechtsteiner EA. National Ambulatory Medical Care
  Survey 2006 Summary.
• National health statistics reports no.3.
  Hyattsville, MD National Center for Health
  Statistics. 2008.

16
NAMCS sample of Community Health Centers
limitations

• 2006 NAMCS PUF only includes CHC physician visits
  
• Additional level of sampling for CHC providers
  increases sampling variability of estimates
• CHC physician visits insufficient for detailed
  analysis of CHC physicians
• 2006-07 CHC PUF file planned for release in 2009
  will include visits to mid-level providers

17
NAMCS/NHAMCS provider-level estimates
18
Physician weight released on NAMCS PUF file

• NAMCS physician weight (PHYSWT) first released on
  2005 PUF
• PHYSWT only on first visit record for physician
• Physician file created by selecting records with
  PHYSWTgt0
• Survey design variables same for physicians as
  visits

19
Physician characteristics on 2006 NAMCS PUF file

• Physician characteristics on PUF
• Physician specialty (SPECR)
• Physician specialty group (SPECCAT)
• Geographic region (REGION)
• Metropolitan statistical area (MSA)
• Solo practice (SOLO)
• Other Induction interview variables on pages
  62-73 of NAMCS PUF documentation

20
Other information on NAMCS Physician weight

• Selected physician estimates presented on page 88
  of 2006 NAMCS PUF documentation
• See pages 27-28 for additional information about
  the physician-level weight

21
Exercise compare visit estimates with physician
estimates

• Compare number of visits by physician specialty
  with number of physicians by specialty
• Steps
• Read NAMCS PUF
• Estimate visits using PUF
• Estimate physicians from physician file

22
Run Exercise 1

• Reads NAMCS PUF and produces weighted frequency
  of visits by physician specialty

23
Output
24

• Run Exercise 2 Creates physician file and
  produces weighted frequency of physicians by
  specialty
• PHYSWTgt0 cases n1,268

25
Output
26

• Run Exercise 3 Compute standard errors of
  physician percentages by specialty using SASs
  PROC SURVEYFREQ

27
Output
28
Physician weight caveatNAMCS PUF files

• PUF physician estimates may differ slightly from
  published physician estimates (e.g. Physicians
  using electronic medical records in 2005 EStat
  report)
• 2005 NAMCS PUF includes only physicians with
  visit records (n1,058)
• EStat estimates include additional 223 in-scope
  physicians unavailable during sample week (on
  vacation or conferences) who responded to
  Physician Induction Interview (n1,281)

29
Provider weights released on NHAMCS PUF file

• Hospital ED weight (EDWT) only on first ED visit
  record for department within sample hospital
• Hospital OPD weight (OPDWT) only on first OPD
  visit record for that department within sample
  hospital
• Create hospital file by selecting records with
  EDWTgt0 or OPDWTgt0 for more accurate variance
  estimates use subpopulation option to select
  either ED or OPD data
• Survey design variables same for hospital
  departments as visits

30
Provider weights released on NHAMCS PUF file
(cont.)

• Selected ED estimates (n364) presented on page
  112 of 2006 NHAMCS PUF documentation
• Selected OPD estimates (n235) presented page
  116-117 of 2006 NHAMCS PUF documentation
• See pages 23-24 for more details on use of ED and
  OPD weight

31
Provider weights released on 2006 NHAMCS PUF file
(cont.)

• ED characteristics on PUF
• Hospital ownership (OWNER),
• Receipt of Medicaid Disproportionate Share
  Program funds (MDSP),
• Receipt of bioterrorism hospital preparedness
  funding (BIOTER),
• Geographic region (REGION),
• Metropolitan statistical area (MSA), and
• Multiple variables on ED use of electronic
  medical records

32
Provider weights released on 2006 NHAMCS PUF file
(cont.)

• OPD characteristics on PUF
• Hospital ownership (OWNER),
• Receipt of Medicaid Disproportionate Share
  Program funds (MDSP),
• Receipt of bioterrorism hospital preparedness
  funding (BIOTER),
• Geographic region (REGION),
• Metropolitan statistical area (MSA), and
• Multiple variables on OPD use of electronic
  medical records

33
Aggregating visit statistics at the physician or
facility level
34
Why aggregate visit data to provider level

• Provides additional information about provider
• Visit characteristic linked to providers can be
  compared across providers
• Examples
• Average caseload by expected payment source
  across EDs
• Average visit duration in EDs by ED visit volume

35
Example

• Note Plus sign indicates median percentages
  across all emergency departments.
• Box represents the middle 50 percent of emergency
  departments.
• Lines represent emergency departments with
  extreme percentages.
• SOURCE Burt, McCaig. Staffing, Capacity, and
  ambulance diversion in emergency department
• United States, 2003-04. Advance data from vital
  and health statistics no. 376. 2006.

36
Steps

• Convert dichotomous analytic variables to 0/1
  format (requires conversion to percentages
  afterwards)
• Convert missing values on continuous variables to
  .
• Use PROC SUMMARY to create one record per
  provider along with aggregate statistic for that
  provider
• Run weighted average on provider file

37
Aggregate ED waiting time from visit file and
estimate distribution across EDs by MSA status

• Run Exercise 4 Read ED visit file and aggregate
  waiting time print first 10 observations

38
Output
39
Aggregate ED waiting time from visit file and
estimate distribution across EDs by MSA status
(Cont.)

• Run Exercise 5 Computes average waiting times in
  hospital EDs in MSAs and Non-MSAs

40
Output for MSAs
41
Histogram and Box plot for MSAs
42
Normal probability plot for MSAs
43
Histogram and Box plot for MSAs
44
Output for Non-MSAs
45
Histogram and Box plot for Non-MSAs
46
Normal probability plot for Non-MSAs
47
Distribution of average waiting time across EDs
in MSAs and Non-MSAs
MSA
Non-MSA
Percentile
48
NAMCS/NHAMCS patient-level estimates
49
Advantages limitations of population-based
surveys

• Population-based surveys
• Estimate persons, including those who never saw a
  health care provider during reference period
  (e.g., last 12 months)
• Health care utilization data subject to recall or
  proxy reporting for children
• Less likely to measure rare medical conditions

50
Advantages limitations of encounter-based
surveys

• Encounter-based surveys
• Estimate the number, kind, and characteristics of
  health care encounters
• Useful in estimating the burden of illness on the
  health care system
• Can estimate rare medical conditions
• Characteristics not subject to recall since
  information found in medical record
• Estimate visits not patients

51
Advantages of translating NAMCS/NHAMCS encounter
data to patient estimates

• Describes patterns of care by frequency of visits
  to the doctor
• Provides more information about patients from
  encounter-level data
• Better describes quality of care to patients vs.
  describing content of encounter

52
How are patients estimated from ambulatory
encounter data?

• Based on multiplicity estimator component of
  network theory
• Multiplicity inherent in ambulatory data
• On average, patients see their physician about 3
  times a year
• Some patients see multiple physicians during year

53
References

• Burt CW and Hing E. Making patient-level
  estimates from medical encounter records using a
  multiplicity estimator. Stat Med 2007
  261762-1774.
• Sirken MG. Network Sampling. In Encyclopedia of
  Biostatistics, Armitage P, Colton T (eds). Wiley
  West Sussex. 1998 2977-2986.
• Birnbaum ZW, Sirken MG. Design of Sample Surveys
  to Estimate the Prevalence of Rare Diseases.
  Vital and Health Statistics, PHS Publication No.
  1, Series 2 (1). U.S. Government Printing Office
  Washington, 1965.

54
Multiplicity of patient visits to physician
V V V V
v v
V
55
Probability of selecting visit increases with
number of patient visits
V V V V
v v
V
4/7
2/7
1/7
56
To count patient only once, adjust visit
probability
V V V V
V
v v
1/1
1/4
1/2
57
Patients estimated using multiplicity estimator
(visit weight)jk
patient weight

Sjk
Number of visits in the past 12 months to sampled
provider
58
Assumptions of patient estimate

• Patient is relation between person and sampled
  doctor
• Assumes previous visits by same patient have
  similar visit characteristics
• One person can be different patients to different
  doctors

59
Limitations of patient estimator

• Assumption of similar characteristics is not
  applicable to all analytical variables
• Patient estimates not equivalent to person-level
  estimates (doesnt count persons with no medical
  encounters)
• Patient estimates limited to physician offices
  and hospital outpatient departments
• Multiplicity information first collected in half
  samples of 2001 NAMCS and NHAMCS (OPD)
• Question on multiplicity of visits available on
  PUF since 2002
• Multiplicity information will be available for ED
  visits in 2007

60
Comparison of distributions for visits and
patients NAMCS 2001
Visits
Patients
61
Percent distribution for people making any health
care visits by number of visits made in one year
NHIS, 1999-2000
Percent of persons
60
50
40
30
20
10
0
1-3
4-9
10
Number of visits
Rate of persons making no health care visit was
17.5.
62
Estimated Percentage of Patients Aged gt45 Years
Who Received Exercise Counseling from their
Primary-Care Physicians, by Sex and Age
GroupNational Ambulatory Medical Care Survey
and National Hospital Ambulatory Medical Care
Survey, United States, 2003-2005
SOURCE Cherry, DK. QuickStat MMWR. November 2,
2007/ 56(43) 1142.
63
Patient weight summary

• Visit records may be re-weighted to provide
  patient-level estimates
• Re-weighted distribution more closely resembles
  population-based estimates
• No change in sampling variance estimation
  procedure other than using the new weight
• Past visits items provide depth to analysis of
  ambulatory care utilization

64
Exercise Compare Visit and Patient estimates
65
Multiplicity measure
66
Creation of a re-weighting factor
Item categories Annual visits Sjk (Interval midpoint) VR (visit ratio)
New 1 1 1
0 visits 1 1 1
1-2 visits 2-3 2.5 .4
3-5 visits 4-6 5 .2
6 visits 7 8 .125
67
Patients estimated using multiplicity estimator
(visit weight)jk
patient weight

Sjk
Number of visits in the past 12 months to sampled
provider
68
SAS code-multiplicity estimator
if pastvis8 then vr1 else if pastvis1 then
vr1 else if pastvis2 then vr.4 else if
pastvis3 then vr.2 else if pastvis4 then
vr.125 vrpatwtpatwtvr
69
Patient estimate exercise

• Compare distribution of visits and patients with
  7 visits during past 12 months by patient age
• Run exercise 6 Computes distribution of visits
  by age

70
Output
71
Patient estimate exercise

• Run exercise 7 Computes distribution of patients
  with 7 visits during past 12 months
• Use patient weight (VRPATWT)

72
Output
73
Number of visits and patients with 7 visits
during past 12 months
74
We hope the topics covered in this session will
be useful to you in future analyses of NAMCS and
NHAMCS data.

• Thank you for attending this session.