Calculation of summary statistics for the SF-36 Health Survey Instrument ------------------------------------------------------------------------
^sf36^ [^if^ exp] [^in^ range] ^, natver^(country abbrev) [^stub^(variable prefix) ^by^(groupvars) ^check^ | ^nocheck^ ^var^lab | ^novar^lab ^val^lab | ^noval^lab ]
where:
(i) ^if^ and ^in^ define valid observations for the command.
(ii) ^natver^ requires a string argument specifying the country in which the data were collected. This is necessary if you wish to calculate the ^SUMMARY PHYSICAL HEALTH^ and ^SUMMARY MENTAL^ ^HEALTH^ measures in addition to the 8 "traditional" SF-36 measures. The two Summary measures require intermediate data (means, standard deviations, and factor scores) specific to each country and usually derived from large national surveys. The allowed ^country abbrev^s are: ^AUS US UK^.
(iii) ^stub^ requires a ^variable prefix^ to be specified. This is the common prefix of the 36 variables. If not specified ^sf36^ assumes that your variables have the prefix ^SF36_^. The prefix is a string whose length must not exceed 5.
(iv) ^check^ implements "manual" range checking. Out-of range data cause informative messages to be displayed, with the current observation number and value of each out-of-range value. The program then terminates, giving the user the opportunity to check and edit data before reinvoking the program. ^nocheck^ (default, equivalent to not specifying this option) implements "automatic" data recoding if out-of-range data are detected. Out-of-range data are set to missing and ^sf36^ continues execution.
(v) ^var^lab specifies that the ^original^ 36 variables are given variable labels defined within the program. Pre-existing labels in the data are overwritten. These labels follow the documentation of the SF-36 Australian version. The setting of ^natver^ does not alter this. ^novar^lab (default, equivalent to not specifying this option) prevents any new variable labels being applied to the data.
(vi) ^val^lab specifies that the ^original^ 36 variables are given value labels defined within the program. Pre-existing labels in the data are overwritten. These labels follow the documentation of the SF-36 Australian version. The setting of ^natver^ does not alter this. ^noval^lab (default, equivalent to not specifying this option) prevents any new value labels being applied to the original variables.
New variables generated by ^sf36^ are given pre-defined variable and value labels regardless of the status of the ^var^lab and ^val^lab options.
Description -----------
^sf36^ is a program for processing SF-36 data using Stata. The SF-36 is a general-purpose health survey instrument published by the Medical Outcomes Trust. It is comprised of 36 questions in 8 domains. In addition, if national weighting data are available, two summary measures (PHYSICAL HEALTH and MENTAL HEALTH) may be calculated.
What you need: -------------
You need a Stata data set with 36 variables corresponding to the 36 questions of the SF-36. You may, of course, have additional variables, for example, a case identifier and other demographic variables.
The 36 SF-36 question variables ^must^ be named in accordance with the following rule:
Variables in the data set must be named ^<stub>nn$^ where:
^<stub>^ is the common variable prefix specified as the argument of the ^stub^ option; if not specified, "SF36_" is assumed; ^nn^ is the question number given as 2 digits (01 to 11); and ^$^ is the question part, if any, for example, a b c etc....
Examples: SF36_03j SF36_05a quest04c quest09b SFQ11a SFQ10
Important note: The UK version is based on "The UK SF36: An analysis -------------- and interpretation manual" Jenkinson et al, Health Services Research Unit, Oxford 1996. This version changes the original numbering of the US SF36 such that:
US UK -- -- question 10 ==> question 9j questions 11a-d ==> questions 10a-d
Also a goodly number of changes are made to the scoring algorithm (cf. MOT SF36 Health Survey Scoring Manual). All these changes are dealt with in ^sf36.ado^. However, ^sf36.ado^ expects that UK data sets already have data scored as in pages 49-51 of Jenkinson. Unlike the US and Australian case, ^sf36.ado^ does not do any recoding. This is because the UK SF36 instrument does not have specified raw data values associated with responses, but simply ticked boxes which are presumably scored according to Jenkinson's algorithm at the time of initial data entry.
What you get: (assume the stub is the default "SF36_") ------------
First, ^sf36^ implements range checking according to rules specified in the SF-36 Reference and Scoring Manual. (See the ^check^ option).
Next, depending on the status of the ^var^lab and ^val^lab options the program labels the 36 original variables. It does so according to the Australian version of the SF-36. (There are trivial semantic differences between countries.)
Value labels are named as ^lab_nn(nn)^, for example, lab_11 is the common value label for 4 variables SF36_11(a,b,c,d), and lab_0405 is the common value label for 7 variables SF36_04(a,b,c,d) and SF36_05(a,b,c).
Then the program handles missing data using either means of non-missing items or by coding the entire scale as missing for that subject as prescribed in the SF-36 documentation.
57 new variables appear in the data set (the original SF36_nn$ variables are untouched). The 57 comprise:
36 new variables (named ^tx_nn$^) represent the original data with missing data imputed by mean substitutions or by missing values substituted for each item in a scale (as specified in the SF-36 docs).
8 new variables represent the untransformed scale scores.
1 new variable for the Health Transition scale (a copy of SF36_02)
8 new variables represent the corresponding transformed scores.
4 new variables associated with the PHYSICAL HEALTH and MENTAL HEALTH SUMMARY scores. These are:
2 new variables representing the raw factor scores: raw_PHYS, raw_MENT (also called AGGregated scores in the SF-36 docs.) 2 new variables representing the standardised scores: std_PHY, std_MENT (also called the transformed scores, PCS and MCS, in the SF-36 docs)
Calculation of these last 4 variables requires nation-specific data hard-coded into ^sf36.ado^. See discussion of national versions below.
The new variables are:
tx_nn$ (nn = 01 to 11; $ = a...j as appropriate)
PF Physical Functioning RP Role Functioning Physical RE Role Functioning Emotional MH Mental Health VT Vitality BP Bodily Pain GH General Health SF Social Functioning
HT reported Health Transition
PF_trans Physical Function-transformed RP_trans Role Physical-transformed RE_trans Role Emotional-tranformed MH_trans Mental Health-transformed VT_trans Vitality-transformed BP_trans Bodily Pain-transformed GH_trans General Health-transformed SF_trans Social Function-transformed
raw_PHYS raw factor score (aggregated) for Physical Health raw_MENT raw factor score (aggregated) for MentalHealth std_PHYS Summary standardised Physical Health/PCS std_MENT Summary standardised Mental Health/MCS
^sf36^ issues three ^summarize^ commands: the first gives summary statistics for the untransformed scales, the second gives them for the transformed scales, and the third gives them for the 4 physical and mental health summary variables (if data are available).
Finally, the program writes a ^note^ to the dataset in memory warning that new variables generated by ^sf36^ were done so according to the current ^if^ and ^in^ conditions. This warning is echoed to the display. The original 36 variables are unchanged.
National data -------------
US data are documented in "SF-36 Physical & Mental Health Summary Scales: A user's manual" Ware, Health Assessment Lab 1994.
Australian data are from the 1995 National Health Survey, Australian Bureau of Statistics.
UK data are from the OHLS as documented in "The UK SF36: an analysis and interpretation manual" Jenkinson et al, Dept Primaty Care and Public Health Oxford 1996 (and errata page).
Examples --------
Variables have prefix "SF36_"; out-of-range values handled manually; Australian data to be used for PHYSICAL and MENTAL HEALTH summary statistics; program supplies variable labels and value labels; summary statistics are produced for all data combined and for each of the states of the Commonwealth:
. ^sf36^ ^using^ mydata, ^check^ ^var^ ^val^ ^natver^(AUS) ^by^(state)
Variables have prefix "sf36Q"; out-of-range values handled automatically; US data to be used for PHYSICAL and MENTAL HEALTH summary statistics; program does not label original 36 variables; new variables and summary statistics only on males:
. ^sf36^ ^using^ mysf36 ^if^ sex=="M", ^stub(^sf36Q^)^ ^natver(^US^)^
Variables have prefix "SF36_"; out-of-range values handled automatically; program supplies value labels, but not variable labels; Australian data used for Summary scales:
. ^sf36^ ^using^ sf36test, ^nocheck^ ^val^ ^natver^(AUS)
(screen output now follows:)
^Note^ ^----^ All new variables in the data set were constructed under the qualifiers: if: ^no if condition^ in: ^no in condition^ Observations on the new variables not satisfying these criteria will be set to missing. Observations on original ^SF36_nn$^ vars are unchanged. A ^note^ to this effect has been written to the data in memory.
^Raw scale statistics^ ^--------------------^
Variable | Obs Mean Std. Dev. Min Max ---------+----------------------------------------------------- PF | 99 25.15869 5.01538 11 30 RP | 93 6.319032 1.621805 4 8 BP | 98 8.95 2.508881 2 12 GH | 96 16.91187 4.572607 6 25 VT | 100 15.25 3.851968 7 23 SF | 100 8.68 1.979083 3 10 RE | 92 5.173913 1.173037 3 6 MH | 100 23.5975 4.876492 7 30 HT | 96 2.53125 .9509344 1 4
^Transformed scale statistics^ ^----------------------------^
Variable | Obs Mean Std. Dev. Min Max ---------+----------------------------------------------------- PF_trans | 99 75.79343 25.0769 5 100 RP_trans | 93 57.97581 40.54513 0 100 BP_trans | 98 69.5 25.08881 0 100 GH_trans | 96 59.55937 22.86304 5 100 VT_trans | 100 56.25 19.25984 15 95 SF_trans | 100 83.5 24.73853 12.5 100 RE_trans | 92 72.46377 39.10123 0 100 MH_trans | 100 74.39 19.50597 8 100
Summary PHYSICAL & MENTAL HEALTH scale statistics (AUS) raw factor scores and standardised scores -------------------------------------------------------
Variable | Obs Mean Std. Dev. Min Max ---------+----------------------------------------------------- raw_PHYS | 88 -.4216362 1.106496 -3.286022 1.81991 raw_MENT | 88 -.1788276 1.118642 -4.008066 1.352772 std_PHYS | 88 45.78364 11.06496 17.13978 68.1991 std_MENT | 88 48.21172 11.18642 9.919338 63.52772
Author ------
Philip Ryan Department of Public Health, University of Adelaide, South Australia pryan@@medicine.adelaide.edu.au