2.0 Methods

2.1 Study Design

The HMCS CHICOUTIMI Health Surveillance Study was a retrospective cohort study that used existing Canadian Armed Forces (CAF) medical records to develop subject-specific data.  Medical and ancillary data included for consideration in the study covered the time period from a subject’s enrolment in the CAF through to 31 December 2009 or the subject’s release date if earlier.

2.2 Subjects

The 250 subjects enrolled in the study were active serving male members on the date 5 October 2004.  There were three cohorts in the study:

  • the HMCS CHICOUTIMI crew (CC, n = 56);
  • the Faslane Care & Custody Team (CCT, n = 42); and
  • the unexposed control submariners (Controls,  n = 152).

Subject selection is summarized in Figure 1. 


There were 56 surviving CC on board the HMCS CHICOUTIMI after the fire, and all were included in the study.

2.2.2 Faslane Care & Custody Team

While the study was being planned, a request was made to include the 42 members of the CCT as a population of interest.  Members of this team worked in the smoke-damaged HMCS CHICOUTIMI after it returned to port in Faslane, Scotland.  While aboard the submarine, they wore a personal protective ensemble.  The list of these individuals was provided to the Occupational Medicine Specialist (OHS) by the leadership of the CCT.

2.2.3 Control Submariners

The Controls were selected as a non-exposed comparison group.  To identify individuals for possible inclusion, DND Human Resources Information Centre (HRIC) provided a list of submariners who satisfied the following inclusion criteria:

  • male;
  • completed the CAF Submarine Basic Qualification Course (AILS) prior to 5 October 2004; and
  • were actively serving as a Regular Member in the CAF as of 5 October 2004.

Personnel in the CC and CCT were then removed from this list.  There were 287 remaining eligible participants, of which 168 were randomly selected to achieve a 3:1 ratio with the CC.

The Personnel Access Support System Verification of Former Service (PASS-VFS) detailed report was reviewed for each selected submariner to confirm that they were actively serving as of 5 October 2004.  Twenty-two were not and were removed from this cohort, to be replaced by 22 individuals randomly selected from the remaining 119 eligible subjects.  All of these members were serving as of 5 October 2004.

Upon completion of the data collection, the data fields related to sick leave (SL) and medical employment limitation(s) (MEL(s)) were reviewed.  Fifteen of the 168 Controls were either on SL or had a MEL that included “unfit sub” or “unfit sea”. The intent was for the Controls to be comprised of submariners who could have been on HMCS CHICOUTIMI at the time of the fire, but were not posted to that boat.  Being on SL or having a MEL that precluded service on submarines would have made these 15 submariners dissimilar to the CC in this regard (all of whom were fit to go to sea as of 5 October 2004).  Hence, it was decided to exclude these individuals from further analyses.  Additionally, one Control subject was subsequently found to have a 2003 release date, and was excluded from the analyses, leaving a final number of 152 Controls.

Figure 1: Control submariners' selection process.  A total of 250 subjects made up the study groups.  HRIC = Human Resources Information Centre.

2.3 Data sources

2.3.1 Medical Records

During the period of data collection, up to 31 December 2009, the CAF was transitioning to an electronic medical record, Canadian Forces Health Information System (CFHIS).  However, as the CFHIS did not, at this time, contain information in addition to what was available in paper records, it was not used.  Thus, the primary source of medical information was the paper-based medical record (CF2034) for each member.  The information found in the CF2034 was supplemented by additional sources as described below.

2.3.2 PASS-VFS

For each subject, a PASS-VFS detailed report was generated via online access to the DND PASS-VFS database.  This report was used to extract and enter into the study data collection form:

  • Date of birth;
  • CAF enrolment date;
  • CAF release date (if released);
  • Release item (if released);
  • Rank; and
  • Posting history.

2.3.3 Pharmacy Profile

A pharmacy profile was generated for each subject by D Med Pol personnel.  It summarized all medications dispensed from CAF pharmacies to the subject, and included the ATC Level 1 (categorical information for the medication, e.g., cardiovascular system, respiratory system, etc.), medication name, and medication dosage.  Dispensed medications were cross-referenced against date dispensed, with results divided into a pre-fire time period (the year preceding 5 October 2004) and a post-fire time period (one-year intervals from 5 October 2004 through to 31 December 2009). 

2.3.4 HMCS CHICOUTIMI Fire Board of Inquiry

The HMCS CHICOUTIMI Fire Board of Inquiry report provided crewmember locations at the time of the onset of the fire (Annex F:  Location of crew when fire erupts, in the Board of Inquiry report).  This information was summarized into a spreadsheet with the following information for each of the CC:  deck level (i.e., 1-deck, 2-deck), fore/aft location (i.e., aft of bulkhead 56, between bulkhead 56 and bulkhead 35), and specific compartment (i.e., control room, bunk space).

2.3.5 Faslane Care & Custody Team Arrival and Departure dates

The CCT list also included the dates that team members arrived in and departed from Faslane.  These data were included in the study data collection form (discussed below). 

2.4 Data Collection Form

Data from the above-listed sources were entered into an interactive PDF form.  The form contained various types of data fields:  date, numeric, text, drop-down lists, and radio buttons. See Annex A for a complete list of the data fields in the form.  All completed forms were provided to the OHS, who extracted data into a spreadsheet for analysis.

The form was divided into three main sections:

  • Baseline:  demographic and medical information for all subjects from enrolment in the Regular Force until the date of the HMCS CHICOUTIMI fire (5 October 2004);
  • CC, immediate fire aftermath:  information related to the fire on board HMCS CHICOUTIMI up until the end of crewmembers’ initial medical assessment in Faslane, Scotland; and
  • Post-fire:  demographic and medical information for all subjects from 5 October 2004 until 31 December 2009 (or the date of release from the CAF).

2.5 File Review process

2.5.1 Overview

Medical records and ancillary information (described in section 2.3) were reviewed and extracted by nurses who entered data into the collection forms.  Three nurses served as file reviewers over the course of the data collection phase. 

2.5.2 File Reviewer Training

Each file reviewer received training on the use of the data collection form, including the type of information to be collected and how it was to be entered into the form.  They were also provided with an instruction manual, which they read before commencing reviews.  The manual provided instructions on how to complete the data fields, inclusion and exclusion criteria, decision rules, and examples.

Reviewers started reviewing files under the direct oversight of the OHS.  Once the reviewers developed sufficient competency, they reviewed files without direct oversight.  The initial forms that were completed “unsupervised” were reviewed separately by the OHS, who compared the form contents to the source documentation in order to verify accuracy and completeness.  Any errors or omissions in the file reviewer’s data collection form were corrected by the OHS.  Feedback was provided to the reviewers as necessary. 

Once satisfied with the reviewer’s performance, the OHS ceased performing the thorough “double-check” of the file reviewer’s work.  At this point, they were considered “trained” and would continue with the data collection work independently.  To reach the “trained” status, file reviewers needed to complete approximately ten data collection forms (i.e., the review of medical and ancillary documentation of ten subjects).

Throughout the course of the data collection phase, the work of the research staff was reviewed by the OHS in order to verify accuracy and completeness.  This was accomplished through the use of “validation files”, which are described in more detail in section 2.5.4.

2.5.3 Medical Records Review Procedure

Before commencing data collection, the OHS prepared the following documentation for use by the file reviewers:

  • PASS-VFS detailed report for each subject;
  • pharmacy profile for each subject;
  • CC location spreadsheet;
  • CCT list with arrival and departure dates; and
  • a master list, containing subject name, Service Number, and a unique six-digit subject study identification number.

Medical records were ordered in small batches from their source locations and delivered to CF H Svcs Gp HQ, where the review work was done.  Subjects were distinguished on the data collection form by their unique six-digit subject study identification number.  Data collection forms did not contain subject name or Service Number and were therefore “anonymized”. 

Due to the nature of the contents of the medical records and the need to abstract specific information regarding the HMCS CHICOUTIMI fire events, file reviewers could not be blinded to subject status (i.e., CC, Controls, or CCT). 

2.5.4 Validation Files

To assess inter-reviewer agreement and to monitor file reviewer performance, 5% of subject files (13 subjects in total) were used as validation files.   

Each validation file was reviewed by all three file reviewers and the OHS.  After the first file reviewer had completed the review of the subject file, it would then be identified as a validation file (the initial reviewer was unaware that the file they were reviewing was going to be used for validation purposes).  The same file would then be reviewed by the remaining two reviewers as per any other subject file, with completed data collection forms sent to the OHS.  Although the first reviewer was blinded to the fact that the data collection form they were completing would be evaluated, the second and third reviewers were not blinded.

After all three reviewers had processed the validation file, the OHS reviewed the validation subject’s medical records, any ancillary information, and then completed a data collection form.  The OHS’s form served as the standard for comparison to the data collection forms completed by the three file reviewers for the same subject.  The performance of the three file reviewers, as compared to each other and the standard was evaluated by the OHS.  Identified inaccuracies or omissions were cross-referenced against source documents to verify the assessment.  The OHS provided feedback to individual reviewers regarding their performance after the completion of each validation file.

Validation files were randomly selected over the course of the data collection phase and included five CC, six Controls, and two CCT.  Two file reviewers were the initial reviewer on four occasions, and one file reviewer was the initial reviewer on five occasions.

With all reviewers and all validation files considered, the average file reviewer accuracy was 93.3% (range:  87.1% to 98.8%).  With respect to inter-reviewer agreement:  complete agreement (for a particular data field, all three file reviewers entered the same information) occurred 73.1% (range:  58.3% to 86.1%) of the time; two-thirds agreement (for a particular data field, two file reviewers entered the same information while the third file reviewer entered different information) occurred 26.1% (range:  14.0% to 41.4%) of the time; and no agreement (for a particular data field, all three file reviewers entered different information) occurred 0.8% (range:  0% to 2.3%) of the time.  Further details regarding the methods and results of the validation file analysis can be found in Annex B.

2.5.5 Data Collection Timelines

Medical and ancillary record review and data abstraction started 25 August 2010 and was completed on 9 January 2012. 

Reviewer 1 worked from 25 August 2010 to 16 December 2011 and completed data collection forms for 101 subjects (not including validation files).

Reviewer 2 worked from 25 January 2011 to 22 December 2011 and completed data collection forms for 67 subjects (not including validation files).

Reviewer 3 worked from 4 April 2011 to 9 January 2012 and completed data collection forms for 58 subjects (not including validation files).

The OHS completed data collection forms on 24 subjects including those used for validation purposes.

When all work days are considered, and work absences due to vacation or illness are included, file reviewers completed an average of 1.5 to 1.8 subject data collection forms per week (including validation files). 

2.6 Data Processing and Analysis

2.6.1 Demographic Information

Enrolment and release dates were based on CAF Regular Force service.  Reserve service before enrolment in the Regular Force was not included in calculations of years of service.  The vast majority of subjects who released from the Regular Force transferred to the Reserve Force or the Supplementary Reserve.  For most subjects, medical records were only available up until the date of release from the Regular Force.  Three Controls, however, had continued postings and medical records post Regular Force release while serving as Reservists.  For these three subjects, post-release service was counted towards their total Regular Force service duration.  These subjects were classified as “currently serving” despite their Regular Force release.

Subjects actively serving as of 31 December 2009 in the Regular Force were classified as “currently serving”.  Subjects, who had released from the Regular Force (with the exception of the three Controls described above), were classified as “released”.

For comparison between groups, dates (e.g., date of birth, enrolment date, etc.) were expressed as the year with a two-digit decimal, with the numbers to the right of the decimal representing the proportion of a complete year.  Rank was categorized as either Junior NCO (Able Seaman/Private to Master Seaman/Master Corporal), Senior NCO (Petty Officer 2nd Class/Sergeant to Chief Petty Officer 1st Class/Chief Warrant Officer), or officer (Naval Cadet/Officer Cadet to Captain(N)/Colonel).

Subject posting history was categorized as either “submarine,” “ship”, or “other” by the OHS.  The amount of time spent at sea could not be determined from the available data.  Submarine postings included Victoria and Oberon-class submarines.

Release items were categorized as “medical release” (Release item 3A or 3B) or “other” (the only non-medical release designations were:  4A, 4B, 4C, 5A, and 5C).

Body Mass Index (BMI) was calculated by dividing a subject’s weight in kilograms by the square of their height in metres.  BMI was categorized as follows:  normal weight (BMI = 18.5 to 24.9), overweight (BMI = 25.0 to 29.9), and obese (BMI = 30.0 or greater).

2.6.2 Sick Leave and Medical Employment Limitations

SL variables captured included:  the SL reason (i.e., diagnosis or medical procedure), diagnostic category (selected from a drop-down list), SL start date, and SL duration in days.  Temporary (Temp) MEL variables included:  limitation description (e.g., “light-duties,” “unfit sea,” “no heavy lifting”, etc.) and reason for limitation (i.e., diagnosis or medical procedure), start date of TempMEL, and TempMEL duration in days.  Permanent (Perm) MEL variables included:  limitation description (e.g., “light-duties”, “unfit sea”, “no heavy lifting”, etc.) and reason for limitation (i.e., diagnosis or medical procedure) and start date of PermMEL. 

Medical record documentation for SL and MELs differed with respect to how diagnosis/reason was recorded.  For SL, the clinician would identify the diagnostic category that necessitated the SL.  In the data collection form, the file reviewer simply had to select the corresponding SL diagnostic category from the drop-down list.  For MELs, however, the clinician would often document all current active diagnoses, and not necessarily identify the specific diagnosis that necessitated the MEL.  In the data collection form, file reviewers abstracted the diagnoses listed in the medical records for the MEL into a text field.  During the analysis, the OHS reviewed the MEL text fields and chose a diagnostic category, often relying on supporting information (e.g., medical history text fields) to make this determination.  The same diagnostic categories were used for both SL and MELs:  cardiovascular; ear, nose, and throat (ENT); endocrine; gastrointestinal; malignancy; musculoskeletal (MSK); neurological; psychiatric; respiratory; urological; other; and, unknown.

Only SL or MELs lasting at least three days were included in the data collection form.  SL and MEL durations were based on calendar days.  SL and MEL information was captured in either the Baseline or Post-Fire sections of the data collection form, depending on the start date of the SL or MEL; start dates that preceded 5 October 2004 were captured in the Baseline section, and start dates of 5 October 2004 or later were captured in the Post-Fire section.

The raw SL and MEL data were processed to create a database.  Baseline information was divided into two time periods:  within five years of the fire date (5 October 1999 to 4 October 2004) and greater than five years before the fire date (earlier than 5 October 1999).  Analyses focused on the baseline “within five years” and post-fire time periods, so that time periods of similar duration pre- and post-fire were compared.  SL or MEL entries that overlapped time periods due to a combination of the start date and duration were split so that the duration was allocated appropriately between time periods and truncated if they went beyond the observation period.  For example:

  • a 30-day SL that started on 30 September 1999 would be recorded in the analysis database as “baseline greater than five years before the fire date, start date 30 September 1999, duration five days”, and “baseline within five years of the fire date, start date 5 October 1999, duration 25 days”; and,
  • a 180-day TempMEL that started on 15 November 2009 would be recorded in the analysis database as “post-fire, start date 15 November 2009, duration 47 days” (duration truncated to the end of the study period, 31 December 2009).

Often, two dates were associated with PermMELs:  the date that the treating clinician sent the subject’s medical records to D Med Pol for review; and the date that D Med Pol officially assigned the PermMEL.  In this circumstance, the earliest possible date was used.  Almost all PermMELs were preceded by at least one TempMEL.  If the TempMEL overlapped with the start date of the PermMEL, then the TempMEL was truncated as if it had ended on the day before the start date of the PermMEL.

For analyses, PermMELs were treated similarly to TempMELs, except that a duration value had to be assigned to the PermMEL.  For post-fire PermMELs, the duration was calculated starting from the start of the PermMEL and ending on the subject’s release date.  Four subjects (two Controls and two CCT) had PermMELs that started before the fire date, and all four had released by 31 December 2009.  These baseline PermMELs were treated similarly to baseline TempMELs that overlapped with the post-fire time period.  For example:

  • a baseline PermMEL with a start date of 24 February 2004 and a subject release date of 22 September 2005, would be recorded in the analysis database as “baseline within five years of fire date, start date 24 February 2004, duration 224 days” (i.e., ending on 4 October 2004), and “post-fire, start date 5 October 2004, duration 352 days” (i.e., ending on the release date).

MELs were classified as follows: 

  • Unfit sea or submarine, and unfit alongside (UFS/UFA):  the MEL description had to state explicitly “unfit alongside” or “unfit work in military environment”;
  • Unfit sea or unfit submarine (UFS):  the MEL description had to state “unfit sea” (with either a statement of fit alongside or no mention of alongside fitness), “unfit submarine,” “no deployment”, “unfit operational environment”, “unfit combat”, “unfit field”, or “draft ashore”;
  • Activity restrictions:  essentially anything else that could have potentially limited day-to-day work activities in some way, such as “light duties”, “unfit dive”, “no drill or PT”, “avoid heavy lifting”, etc., but the subject was still “fit sea” or “fit sub”;
  • Other restrictions:  these included noted limitations that likely would not have directly limited day-to-day work activities, such as “requires medical follow-up every two weeks”;
  • Unknown restrictions:  no information on the type of restriction was provided in the data collection form; and
  • TempMEL descriptions that were unlikely to have an operational impact (e.g., “avoid dehydration”, “requires medical follow-up every 6 months”, “requires maximum hearing protection”, etc.) were not included in the database.

There were instances where SL overlapped with a MEL or a MEL of one type would overlap with another MEL type.  A work restriction hierarchy was developed to avoid double counting days.  It was, in order of severity, SL> UFS/UFA> UFS> activity restrictions> other restrictions.  Where there was overlap, the higher severity work restriction took precedence.  For example:

  • if a subject had an UFS/UFA Temp MEL starting 5 July 2006 for a duration of 90 days and they also had a SL episode starting 1 August 2006 for 30 days, then this would be captured in the database as “SL, start date 1 August 2006, duration 30 days,” and “TempMEL, UFS/UFA, start date 5 July 2006, duration 60 days” (i.e., the SL duration has been subtracted out of the MEL duration); and
  • if a subject had an UFS TempMEL starting 1 January 2007 for 180 days, and also had two activity restriction TempMELs (one starting 30 March 2007 for ten days and another starting 15 June 2007 for 30 days), then this would be captured in the analysis database as “TempMEL, UFS, start date 1 January 2007, duration 180 days”, and “TempMEL, activity restriction, start date 30 June 2007, duration 15 days”.

In some cases, subjects had multiple contiguous TempMELs for the same medical condition.  Frequently, these TempMEL time periods (as recorded in the medical records) would overlap or would not be exactly contiguous (i.e., the first 180-day UFS TempMEL would end five days before the second 180-day UFS TempMEL started).  Instances of TempMEL overlap were treated as described previously.  For instances where long duration TempMELs were not quite contiguous, the duration gap between the TempMELs was assumed to be also affected by a TempMEL (i.e., it was presumed that the MEL gap was an administrative one as opposed to the subject being unfit sea for six months, suddenly completely well for five days, and then unfit sea again for six months).  Examples of how multiple contiguous TempMELs were entered into the analysis database are provided below:

  • if a subject had an UFS/UFA TempMEL starting 1 January 2007 for 180 days and then a subsequent UFS/UFA TempMEL starting 5 July 2007 for 180 days for the same condition, this would be captured in the analysis database as “TempMEL, UFS/UFA, start date 1 January 2007, duration 365 days”; or
  • if a subject had an UFS TempMEL starting 1 March 2006 for 90 days and then a subsequent UFS TempMEL starting 1 May 2006 for 90 days for the same condition, this would be captured in the analysis database as “TempMEL, UFS, start date 1 March 2006, duration 151 days” (i.e., the duration is 1 March 2006 to 30 April 2006 + 90 days).

The variable of interest for the SL and MEL analyses was duration in days.  Duration was divided into diagnostic categories, and also by the type of work restriction (i.e., SL, UFS/UFA, UFS).  For example, a subject who had a 30-day respiratory SL, a 14-day ENT activity restriction MEL, a 60-day respiratory UFS MEL, and a 90-day ENT UFS/UFA MEL could be summarized as follows:

  • total restricted days (SL plus MEL) for respiratory and ENT was 90 days and 104 days, respectively, for an all diagnosis total of 194 days;
  • the total SL days was 30 and total MEL days was 164; and
  • the total days SL or UFS/UFA or UFS was 180.

To account for different lengths of service in the different time periodsFootnote 6, the number of restricted days was divided by the total number of days that the subject served in the CAF during that time period.  In this respect, for the baseline period (5 October 1999 to 4 October 2004) the total possible number of days was 1827; while for the post-fire time period (5 October 2004 to 31 December 2009), it was 1913 days.

Only two subjects had enrolment dates that occurred after 5 October 1999 and their days served was based on enrolment date.  For the post-fire period, the number of days served ranged from 23 days to 1871 days for the 89 subjects who released.  Fourteen subjects released within one year of the fire date (i.e., released prior to 5 October 2005):  three CC, four CCT, and seven Controls.  Using this method, a subject with a specific number of restricted duty days could have a different estimate of proportional loss based on time served.  For example, using 194 as the total restricted days (assuming that this was the post-fire time period), a subject would be restricted for:

  • 53.2% of the time if released on 4 October 2005 (194 / 365 * 100 = 53.2%); and
  • 10.1% of the time if still serving as of 31 December 2009 (194 / 1913 * 100 = 10.1%).

As mentioned above, the OHS reviewed the MEL text fields and assigned an appropriate diagnostic category, often relying on supporting information (e.g., medical history text fields) to determine the primary diagnostic category that necessitated the MEL.  MELs were typically assigned a single diagnostic category, except where a dual diagnosis category was warranted because both conditions were active concurrently.  In this case, the total number of MEL days was divided equally between the two diagnostic categories.

2.6.3 Medical History Text Field Diagnoses

The data form contained text fields that were used for the abstraction of medical history details.  For the baseline and post-fire time periods, separate text fields were used to enter details of any respiratory, ear, nose and throat, cardiovascular, gastrointestinal, psychiatric, neurological, and “other” conditions.  With the exception of the “other” text field, reviewers were instructed to capture information related to any condition that fell within the text field’s diagnostic category.  These included details such as diagnosis, pertinent dates, and additional relevant details related to treatment and condition severity.  This information was found in and extracted from clinic notes (CF2016), consultation letters, and periodic medical evaluation notes.

For conditions other than respiratory, ear, nose and throat, cardiovascular, gastrointestinal, psychiatric, or neurological, the reviewer would judge whether to include information under “other” in the form.  In this respect, the instruction was to exclude minor or self-limited one-time conditions; and to include conditions that were chronic or that occurred frequently, or were a significant isolated event (e.g., surgery, hospitalization, etc.).

The OHS compiled all of the subject medical history text fields across the different diagnostic categories and reviewed the captured information.  The total number of words included in the baseline and post-fire medical history text fields was 113,015 and 80,862, respectively.  Any misclassified diagnoses (e.g., “headaches” documented in the psychiatric text field instead of the neurological text field, or “pneumonia” documented in the “other” text field instead of the respiratory text field, etc.) were assigned to the correct diagnostic category by the OHS.

The OHS extracted information from the compiled post-fire medical history text fields into a database, which included diagnostic category (mirroring the diagnostic list used for SL and MELs), specific diagnosis, and whether or not the diagnosis was present in the baseline time period.  With respect to the latter variable, for each post-fire diagnosis identified, the OHS cross-referenced the subject’s corresponding baseline medical history information (from the date of enrolment to the date of the fire).  If the post-fire diagnosis had not been documented in the baseline time period, then it was coded as a “new” post-fire diagnosis.  If the post-fire diagnosis had been documented in the baseline time period, then it was coded as a pre-existing condition.

The OHS used judgment as to which diagnoses to include in the database.  The intent was to include medical conditions that were significant or chronic/recurrent and to exclude minor self-limited conditions.  For example, if a subject had a single documented headache, this would not be included in the database, but if a subject had recurrent migraine or cluster headaches, then this would be included in the database.  Similarly, one or several self-limited upper respiratory tract infections (i.e., common cold) would not be included in the database, but asthma would be included.

Analysis of the medical history text fields does not necessarily provide a complete picture of all medical conditions present.  Some chronic conditions, such as hyperlipidemia or hypertension, might not have been documented (for various reasons) by clinicians.  Similarly, conditions identified in laboratory reports (e.g., micro-hematuria or proteinuria, etc.) might not have been included in written clinical notes. 

2.7 Statistics

Statistical analyses were restricted to demographic variables and selected health outcomes.  Pair-wise comparisons for continuous variables were done using an independent sample T-test, and categorical variables were compared using a Chi square test.  The threshold for statistical significance was a P value of 0.05.  SPSS software was used for the majority of analyses.  Stata version 11 was used to compute incidence rates and their corresponding confidence intervals as well as the incidence rate ratios (IRR), which were estimated using Poisson’s regression models.

Page details

Date modified: