help for icd9tmpm -------------------------------------------------------------------------------
icd9tmpm -- Trauma Mortality Prediction Model (ICD-9-CM version)
ICDPIC Version 3.0 requires STATA 8.0 or higher. ICDPIC Version 3.0 has been tested in STATA 10 and STATA 11, but the authors believe it should also work without incident in STATA 8 and STATA 9. If you have any problems using ICDPIC Version 3.0 in STATA 8 or STATA 9, please inform the authors.
ICDPIC Version 3.0 may be installed from within STATA using the ssc command. If you installed a previous version of ICDPIC from the SSC archives website using the ssc command, we suggest that you first delete it by typing ssc uninstall icdpic followed by ssc install icdpic. Alternatively, you may use ssc install icdpic, replace. See help for ssc.
If you installed any previous ICDPIC files obtained directly from the authors, please delete them ALL (.ado, .hlp and .dta files) to avoid any conflicts with ICDPIC 3.0 files.
Please enter complete variable names in the ICDPIC Version 3.0 dialog boxes. Do not use abbreviations.
New to Version 3.0 is the addition of a dialog box (.dlg) file associated with each individual ICDPIC Version 3.0 program (.ado) file. To access the ICDPIC dialog box, and all the ICDPIC programs, type: db icdpic. Typing icdpic, as in earlier versions, will still work, but ONLY with icdpic. For example, to access the ICD-9-CM TMPM program directly, type: db icd9tmpm. Typing icd9tmpm, as in previous versions, will produce an error.
Fixed in ICDPIC Version 3.0 is the ability to use path\file names containing spaces.
Fixed in ICDPIC Version 3.0 is the ability to run in STATA 11.0.
Fixed in ICDPIC Version 3.0 is a bug that caused the triss program to crash if the rts variable was named anything other than "rts".
New in ICDPIC Version 3.0 (trauma program only) is the ability to choose whether an AIS value of 6 automatically forces an ISS of 75 or to automatically have all AIS values of 6 changed to an AIS value of 5 and then have the ISS calculated normally.
All dialog boxes in ICDPIC Version 3.0 have memory. Each time a dialog box is opened within the same STATA session, it will remember the values last entered.
All dialog boxes in ICDPIC Version 3.0 have the following buttons:
OK executes the program and removes the dialog box from the screen.
SUBMIT executes the program and leaves the dialog box on the screen. Note that if an error message is generated the dialog box may be minimized.
CANCEL removes the dialog box from the screen and does nothing. Clicking on the close icon of the dialog box does the same thing.
HELP leaves the dialog box on the screen and presents the program help file. The HELP button has a question mark on it.
COPY leaves the dialog box on the screen and copies the program command to the clipboard.
RESET resets the values of the controls in the dialog box to their initial state, just as if the dialog box were invoked for the first time. Each time a user invokes a dialog box, its controls will be filled in with the values the user last entered. RESET restores the control values to their defaults. The RESET button has an R on it.
The ICD-9-CM TMPM (ICD-9-CM Trauma Mortality Prediction Model) dialog box will open. Follow the instructions.
The ICDPIC dialog box will open. Choose ICD-9-CM TMPM and click OK or Submit. The ICD-9-CM TMPM (ICD-9-CM Trauma Mortality Prediction Model) dialog box will open. Follow the instructions.
icd9tmpm estimates patient probability of death,
Pd = normal(TMPM)
TMPM = (C0 + (C1 X I1) + (C2 X I2) + (C3 X I3) + (C4 X I4) + (C5 X I5) + (C6 X > S) + (C7 X I1 X I2)
Pd is the mortality predicted by TMPM and normal is the standard normal distribution function. I1 through I5 are the MARC values ordered with the greatest MARC value (worst injury) first, second greatest MARC value second, up to the fifth worst injury. S is a Boolean value set to 1 if the two worst injuries occur in the same body region and 0 otherwise. I1 X I2 represents the interaction of the worst and second worst injuries a patient has sustained. C0 through C7 are coefficients.
See also the Options and Remarks sections for IMPORTANT information on, and requirements for, icd9tmpm.
ICD-9-CM diagnosis codes in the user's data may, or may not, contain a decimal point. Simply choose the option that applies to your data in the ICD-9-CM TMPM dialog box.
The user's data must contain ICD-9-CM diagnosis codes. Diagnosis codes in the user's data must be of type string. A decimal point in the diagnosis codes is optional. Diagnosis codes should have a width of 5 (6 if a decimal point is present). The diagnosis code prefix must be the same for all diagnosis codes and numbered sequentially starting with 1, for example, dx1...dxN.
icd9tmpm adds the following variables to a new copy of the user's data stored on disk:
marc_1-marc_n: MARC value for Nature of Injury codes (N-Codes) 1..n inj1-inj5: injuries (N-Codes) associated with the top 5 MARC values (5 worst injuries) marc1-marc5: top 5 MARC values (5 worst injuries) high1-high2: ISS body region of the two worst injuries (top two MARC values) respectively same_reg: Boolean variable to indicate if top two injuries occurred in the same body region ICD9TMPM: Trauma Mortality Prediction Model value ICD9_POD: probability of death
Variables ICD9TMPM and ICD9_POD are rounded to the nearest 0.000001.
icd9tmpm requires the use of lookup tables micd9_s1.dta and micd9_s2.dta. These data tables are supplied. _s1 tables are for use with data without the decimal point in ICD-9-CM codes and _s2 tables are for data that include the decimal point. See help for icdpic, specifically the LOOKUP TABLES part of the Remarks section.
NOTES ON METHODOLOGY AND VALIDATION
The Trauma Mortality Prediction Model (ICD-9-CM TMPM) has the same goal as the ISS. Both provide a single number summary of a patient's injuries when those injuries are described in the ICD-9-CM lexicon. However ISS considers only at most three of a patient's injuries, and provides only a relative measure of injury severity (0 to 75). ICD-9-CM TMPM, by contrast, considers as many as 5 of a patient's injuries and results in an actual prediction of that patient's mortality (0% to 100%) based upon the injuries sustained.
Although ISS and ICD-9-CM TMPM have similar goals, the models upon which they are based are quite different: ISS is simply the sum of the squared values of the worst AIS severities in each of the three most severely injured body regions. ICD-9-CM TMPM, by contrast, is based upon a probit model that includes the five worst injuries (regardless of where they occur in the body), as well as terms that describe interactions and whether the two worst injuries are in the same body region. A further difference is the source of information that ISS and ICD-9-CM TMPM use to assign severity to individual injuries: ISS uses the AIS severity created by panels of experts over the years while ICD-9-CM TMPM uses empirically derived values for each ICD-9-CM trauma code. The precise details of this derivation are somewhat mathematical, but the process is basically a regression model that provides a predicted survival for each injury in the ICD-9-CM lexicon expressed as its probit transformation. These "Model Adjusted Regression Coefficients" (MARC values) were derived from 702,229 patients injured between 2001 and 2005 and cared for at trauma centers that provided information to the National Trauma Data Bank.
The details of the mathematics behind the derivation of MARC values and construction and validation of ICD-9-CM TMPM are available. (TMPM-ICD9: A Trauma Mortality Prediction Model Based on ICD-9-CM Codes Glance, LG, Annals of Surgery, in press.) In this study ICD-9-CM TMPM provided predictions that were substantially more accurate, both in terms of discrimination and calibration, than did ISS.
Perhaps adding to the appeal of ICD-9-CM TMPM is the fact that the original version of this model that uses injuries described in the AIS lexicon is also available (AIS TMPM). Because administrative data often describe patient injuries in either the ICD-9-CM or AIS lexicon, but not both, we anticipate the two TMPM programs (ICD-9-CM TMPM and AIS TMPM) will be useful to researchers who perhaps are confronted with data sets in which patients are described in the AIS lexicon in some and described in the ICD-9-CM lexicon in others.
Note that, although ICD-9-CM TMPM uses only anatomic injuries in its predictions, there in nothing precluding the addition of other information (age, GCS, etc.) to outcome predictions. A logistic model predicting outcome for any patient data set can easily be constructed with the anatomic injury component of the model encapsulated in the ICD-9-CM TMPM value. Note, however, that if ICD-9-CM TMPM is to be incorporated into a larger logistic model it should enter such a model as its logit (log(Ps /(1-Ps))) transformation.
David E. Clark, M.D.
Maine Medical Center, Portland, Maine, USA University of Vermont College of Medicine, Burlington, Vermont, USA Harvard Injury Control Research Center, Harvard School of Public Health, Boston, Massachusetts, USA
Turner M. Osler, M.D.
University of Vermont College of Medicine, Burlington, Vermont, USA
Correspondence to Dr. Osler, Department of Surgery University of Vermont 789 Orchard Shore Road Colchester, VT 05446 Email: firstname.lastname@example.org
David R. Hahn
Maine Medical Center, Portland, Maine, USA
Osler T, Glance L, Buzas JS, Mukamel D, Wagner J, Dick A. A Trauma Mortality Prediction Model Based on the Anatomic Injury Scale. Ann Surg 2008;247:1041-1048.
Glance L, Osler T, Mukamel D, Meredith W, Wagner J, Dick A. TMPM-ICD9: A Trauma Mortality Prediction Model Based on ICD-9-CM Codes. Ann Surg In Press.
help for ais_tmpm
help for icdpic