How to understand and use National Ambulatory Medical Care Survey (NAMCS) and National Hospital Ambulatory Medical Care Survey (NHAMCS) data for clinical research
Yuwei Zhu
10-29-2004 Dept of Biostatistics
1
Overview
I. Survey Background II. Survey Methodology III. Technical Considerations IV. Getting the Data – Using Raw Data Files V. Example VI. Data Analysis – SAS, STATA, SUDAAN VII. Other Public Domain Data
2
NAMCS and NHAMCS
Performed by: Centers for Disease Control and Prevention (CDC) National Center for Health Statistics, Division of Health Care Statistics, and National Health Care Survey
3
National Ambulatory Medical Care Survey (NAMCS) History
Survey began in 1973 Annual data collection through 1981 Conducted in 1985 Annual began again in 1989
4
NAMCS
Classified by the American Medical Association and the American Osteopathic Association as delivering “office-based, patient care” Healthcare providers within private, non–hospital-based clinics and health maintenance organizations (HMOs) are within the scope of the survey
5
NAMCS
Patient visits made to the offices of non– federally employed physicians – Excluding: Anesthesiology Radiology Pathology
6
In-Scope NAMCS locations
Freestanding clinic Federally qualified health center Neighborhood and mental health centers Non-federal government clinic Family planning clinic HMO Faculty practice plan Private solo or group practice
7
Out-of-Scope NAMCS locations
Hospital EDs and OPDs Ambulatory surgicenter Institutional setting (schools, prisons) Industrial outpatient facility Federal Government operated clinic Laser vision surgery
8
NAMCS
NAMCS uses a multistage probability sample design to obtain –Primary sampling units (PSUs) –Physician practices within the PSUs –Patient visits within physician practices
9
Sample design - NAMCS
112 PSUs (counties)
– Counties – Groups of counties – County equivalents (such as parishes or independent cities) – Towns – Townships
Nonfederally employed, office-based physicians stratified by specialty, 3,000 physicians About 30 visits per doctor over a randomly selected 1-week period, 25,000 visits
10
National Hospital Ambulatory Medical Care Survey (NHAMCS) History
Survey began in 1992 Annual data collection
11
NHAMCS
National sample of visits to the EDs and outpatient departments of noninstitutional general and short-stay hospitals in the United States Excluded hospitals: – Federal – Military – Veterans Administration
12
NHAMCS
This survey uses a 4-stage probability design with samples –geographically defined areas –hospitals within these areas –clinics within the hospital –patient visits within clinics. The first stage is similar to NAMCS
13
Sample design - NHAMCS
112 PSUs (counties) Panel of 600 non-Federal, general or short stay hospitals Clinics (OPDs) and emergency service areas (EDs), 400 EDs and 250 OPDs About 200 visits per OPD, 100 per ED over random 4-week period, 37,000 ED and 35,000 OPD visits
14
NHAMCS Scope
OPD was intended to be parallel to the NAMCS in the hospital setting
General medicine, surgery, pediatrics, ob/gyn, substance abuse, and “other” clinics are inscope Ancillary services are out of scope
15
Data Items
Patient characteristics
– Age, sex, race, ethnicity
Visit characteristics
– Source of payment, continuity of care, reason for visit, diagnosis, treatment
Provider characteristics
– Physician specialty, hospital ownership…
Drug characteristics added in 1980
– Class, composition, control status, etc.
16
Repeating fields (from text entries)
Up to 3 fields each… – Reason for visit – Physician’s diagnosis – Cause of injury Diagnostic services (6 fields) Surgical procedures (2 fields) Medications (6 fields) – Drug ingredients (5 fields) – Therapeutic class (3 fields – 2002 on)
17
Coding Systems Used
Reason for Visit Classification (NCHS) ICD-9-CM for diagnoses, causes of injury and procedures Drug Classification System (NCHS) National Drug Code Directory
18
Drug Data in NAMCS/ NHAMCS
What is a “Drug Mention” ? Any of up to 6 medications that were ordered, supplied, administered, or continued during the visit.
Respondents are asked to report trade names or generic names only (not dosage, administration, or regimen).
19
Drug Characteristics
Generic Name (for single ingredient drugs) Prescription Status Composition Status Controlled Substance Status Up to 3 NDC Therapeutic Classes (4-digit) Up to 5 Ingredients (for multiple ingredient drugs)
20
Some User Considerations
NAMCS/NHAMCS sample visits, not patients No estimates of incidence or prevalence No state-level estimates Not sampled by setting or by nonphysician providers May capture different types of care for solo vs. group practice physicians
21
Data uses
Understand health care practice Examine the quality of care Track certain conditions Find health disparities Measure Healthy People 2010 objectives Serve as benchmark for states
22
Data users
Over 100 journal publications in last 2 years Medical associations Government agencies Health services researchers University and medical schools Broadcast and print media
23
Sample Weight
Each NAMCS record contains a single weight, which we call Patient Visit Weight Same is true for OPD records and ED records This weight is used for both visits and drug mentions
24
Reliability of Estimates
Estimates should be based on at least 30 sample records AND Estimates with a relative standard error (standard error divided by the estimate) greater than 30 percent are considered unreliable by NCHS standards Both conditions should be met to obtain reliable estimates
25
How Good are the Estimates?
Depends on what you are looking at. In general, OPD estimates tend to be somewhat less reliable than NAMCS and ED. Since 1999, Advance Data reports include standard errors in every table so it is easy to compute confidence intervals around the estimates.
26
Sampling Error
NAMCS and NHAMCS are not simple random samples Clustering effects of visits within the physician’s practice, physician practices within PSUs, clinics within hospitals Must use some method to calculate standard errors for frequencies, percents, and rates
27
Ways to Improve Reliability of Estimates
Combine NAMCS, ED and OPD data to produce ambulatory care visit estimates Combine multiple years of data Aggregate categories of interest into broader groups.
28
NAMCS vs. NHAMCS
Consider what types of settings are best for a particular analysis – Persons of color are more likely to visit OPD's and ED's than physician offices – Persons in some age groups make disproportionately larger shares of visits to ED's than offices and OPD's
29
File Structure
Download data and layout from website http://www.cdc.gov/nchs/about/major/ahcd/ ahcd1.htm Flat ASCII files for each setting and year NAMCS: 1973-2002 NHAMCS: 1992-2002
30
Trend considerations
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+ – 8 medications Diagnostic & therapeutic checkboxes vary Use spreadsheet for significance of trends
31
Example
Hypothesis -- Educational Efforts Targeted
at Judicious Antibiotic Use Will Reduce Prescription Rates in all Treatment Settings
32
Study Design
Retrospective collection of data from – NAMCS – NHAMCS 1994-2000 study years Antibiotic prescribing patterns and diagnoses Children <5 years of age Clinic type -- Pediatric Physician type – Pediatrician or Family Medicine
33
Data Stratification
Race – White, Black and other Time period – 94 & 95, 96 & 97, 98 & 00 Antibiotics – Penicillin's, Cephalosporins,
Erythromycin/lincosamide/macrolides,Tetracyclines, Chloramphenicol derivatives, Aminoglycosides, Sulfonamides and trimethoprim, Miscellaneous antibacterial agents, and Quinolone/derivatives
Diagnoses -- Otitis media, Sinusitis, Pharyngitis,Bronchitis,Upper respiratory tract infection (URI)
34
Overall Antibiotic Rates in Children <5 Based on Source of Care
Rates per 1000 children
2000 1500 1000 500 0 1994 1995 1996 1997 1998 1999 2000 Years Hospital-based ED Office-based
35
Total Care Years
White
Black Rate 95% CI Ratio 3102 1.34 1.22, 1.47* 1.02, 1.08* 0.70, 1.34
Visit rates 1994per 1000 1995 children aged <5 1996years 1997 19982000
4150
4529
4320
1.05
4204
4302
0.98
36
White children
% Distribution health care visit site
100%
Black Children
80% 60% 40%
20%
Hospital-base
ED
Office-based
0% 1994- 1996- 19991995 1998 2000 Years
37
1994- 1996- 19991995 1998 2000
Total Care Antibiotic prescription rates per 1000 children aged <5 years
Years 19941995
19961997
White 1494
Black 998
Rate 95% CI Ratio 1.50 1.48, 1.51*
1.08 0.96, 1.22
1421
1320
19982000
1118
1074
1.04
0.86, 1.24
38
Total Care
Years 19941995
Rate White Black Ratio 816 520 1.57
95% CI 1.46, 1.69*
1.04, 1.07* 0.69, 1.58
Otitis media rates per 19961000 1997 children aged <5 1998years 2000
779
739
1.06
630
603
1.05
39
Results
Decline in antibiotic prescribing in children <5 years; most notable in office-based and emergency department settings Penicillin's were common antibiotics used Most common diagnosis in all three settings was otitis media
Natasha B. Halasa, Marie R. Griffin, Yuwei Zhu, and Kathryn M. Edwards. Difference in antibiotic prescribing patterns for children aged less than five years in the three major outpatient settings, Journal of Pediatrics. 2004; 144:200-205
40
Code to create design variables: survey years 2001 & earlier
CPSUM=PSUM; CSTRATM = STRATM; IF CPSUM IN(1, 2, 3, 4) THEN DO; CPSUM = PROVIDER +100000; CSTRATM = (STRATM*100000) +(1000*(MOD(YEAR,100))) + (SUBFILE*100) + PROSTRAT; END; ELSE CSTRATM = (STRATM*100000);
41
SUDAAN version 8.0.2 example
proc crosstab data=test1 design=WOR filetype=sas; Nest stratm psum subfile prostrat year provider dept su clinic/missunit; Totcnt poppsum _zero_ _zero_ _zero_ popprovm _zero_ popsum _zero_ popvism;
Weight patwt; Tables sex*ager; run;
42
SUDAAN version 8.0.2 example
proc crosstab data=test1 filetype=sas; Nest stratm psum ;
Weight patwt; Tables sex*ager; run;
43
STATA version 8. example
Use http:// ***/test1 svyset [pweight=patwt], strata(cstratm) psu(cpsum)
svytab sex ager svymean age
44
SAS version 9.1 example
proc surveyfreq data=test1; tables sex*ager; strata cstratm; cluster cpsum; weight patwt; run;
45
Some considerations: SUDAAN vs. SAS Proc Surveymeans
SUDAAN •design variables=cstratm, cpsum (1-stage design) •nest=cstratm, cpsum PROC Surveymeans •design variables=cstratm, cpsum (1-stage design) •strata cstratm •cluster cpsum
•Sort by design variables
•Weight data: Patwt
•Sort not needed
•Weight data: Patwt
•Subgroup=identify categorical variables •Tables=analysis variables
•Class=identify categorical variables •Var=analysis variables
46
If nothing else, remember…The Public Use Data File Documentation is YOUR FRIEND!
Each booklet includes: – A description of the survey – Record format – Marginal data (summaries) – Various definitions – Reason for Visit classification codes – Medication & generic names – Therapeutic classes
47
Other Public Domain Data
CDC WONDER -- http://wonder.cdc.gov/ National Center for Health Statistics -http://www.cdc.gov/nchs/ National Health and Nutrition Examination Survey (NHANES) -http://www.cdc.gov/nchs/nhanes.htm National Health Interview Survey (NHIS) -http://www.cdc.gov/nchs/nhis.htm National Survey of Family Growth (NSFG) -http://www.cdc.gov/nchs/nsfg.htm Census -- http://www.census.gov/
48
Other Public Domain Data (cont.)
Dept. of Health, TN http://hitspot.state.tn.us/hitspot/hit/main/ SPOT/frames/SPOT/index.htm
49
Thanks
Natasha Halasha Susan Schappert - National Center for Health Statistics Linda McCaig & David Woodwell National Center for Health Statistics
50
Questions?
51