You are on page 1of 66
Direct Medical Costs of Motorcycle Crashes in Ontario Journal: CMAJ Manuscript ID CMAJ-17-0337.R2 Manuscript

Direct Medical Costs of Motorcycle Crashes in Ontario

Journal:

CMAJ

Manuscript ID

CMAJ-17-0337.R2

Manuscript Type:

Research - Cohort study (retrospective)

Date Submitted by the Author:

n/a

Complete List of Authors:

Pincus, Daniel; University of Toronto,

Wasserstein, David; Sunnybrook Health Sciences Centre

Nathens, Avery; Sunnybrook Health Sciences Center, ; Institute for Clinical

Evaluative Sciences,

Bai, Yu; Institute for Clinical Evaluative Sciences,

Redelmeier, Donald A.; University of Toronto, Medicine

Wodchis, Walter; University of Toronto, Health Policy, Management, and

Evaluation

Keywords:

Epidemiology, Health Policy, Health Economics, Public Health

More Detailed Keywords:

Motorcycles, Traffic Accidents, Direct medical costs

 

Background. No reliable estimate of costs incurred by motorcycle crashes

(MCs) exists. Our objective was to calculate the direct costs of all publicly

funded medical care provided to individuals following MCs compared to

automobile crashes (ACs).

Methods. We conducted a population based, matched cohort study of

adults in Ontario that presented to hospital because of a MC or AC from

2007 through 2013. For each case, we identified one control absent a motor vehicle crash during the study period. Direct costs for each case and control were estimated as 2013 $ Canadian from the payer perspective using methodology that links healthcare use to individuals over time. MC- and AC-attributable costs within 2 years were then calculated using a difference-in-differences approach.

Abstract:

Results. We identified 26,831 patients injured from MCs and 281,826 injured from ACs. Mean MC- and AC-attributable costs were $5,825 and $2,995, respectively (p<0.0001). The rate of injury was triple for MCs compared to ACs (2,194 injured yearly/100,000 registered motorcycles versus 718 injured yearly/100,000 registered automobiles; IRR=3.1, 95% CI=2.8-3.3, p<0.0001). Severe injuries, defined as those with an Abbreviated Injury Scale >=3, were 10 times greater (125 severe injuries per yearly/100,000 registered motorcycles versus 12 severe injuries per yearly/100,000 registered automobiles; IRR=10.4, 95% CI=8.3-13.1,

p<0.0001).

Interpretation. Considering both the attributable cost and higher rate of injury, we found each registered motorcycle in Ontario costs the public

For Peer Review Only

Page 1 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

healthcare system 6 times the amount of each registered automobile. Medical costs may provide an
healthcare system 6 times the amount of each registered automobile.
Medical costs may provide an additional incentive to improve motorcycle
safety.
automobile. Medical costs may provide an additional incentive to improve motorcycle safety. For Peer Review Only

For Peer Review Only

Page 2 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

COMMENTS

1

Dear Editorial Board of The CMAJ,

MC costs

Thank you very much for the opportunity to revise our manuscript "CMAJ-17-0337.R1: Direct

Medical Costs of Motorcycle Crashes in Ontario". We sincerely appreciate the thoughtful

reviews and believe that in addressing the feedback from reviewers we have improved the

paper. Our responses to specific comments from the editorial board and individual reviewers are listed below.

Thank you very much again.

Daniel Pincus on behalf of the authors.

--

Editorial Board:

Comment 1. What sort of motor bikes are covered by your codes? For example are electric-

assisted bicycles in here? Do you have any estimate as to how valid the coding is?

This is an important point. 'Mopeds, motor scooters, and motorized bicycles’ were included in

our definition of ‘motorcycles’ as per the International Classification of Diseases and Related

Health Problems, 10 th Revision published by the Canadian Institute for Health Information (ICD-

10-CA). There is no way to distinguish between these different vehicle subtypes using the ICD-

10-CA to our knowledge.

Including these other motorcycle subtypes may have increased the overall rate of injury

observed in our study. However, we know from the Ontario Road Safety Annual Report (ORSAR)

data that does distinguish between motorcycle subtypes that moped crashes result very few

health care interactions: for example, 25 emergency room visits, 2 hospital admissions, and

zero deaths in 2013.(1) In other words, mopeds result in <1% of the hospital presentations, and

even fewer of the hospital admissions, attributed to motorcycles overall. Including these lower speed vehicles would have also decreased the proportion of overall injuries that were severe and/or resulted in death. For these reasons, it is unlikely our cost estimates per vehicle were affected by this coding limitation.

Although the ICD codes for external causes of injury are considered extremely accurate, those relating to MVCs specifically have not undergone a formal validation to our knowledge. However, their validity is supported by:

1) Coding of ‘Type 9’ diagnoses, such as those used to define motorcycle and automobile

injuries in this study, being a mandatory requirement in the databases used in this

study.(2)

2) The use of ‘Type 9’ codes use in several prior studies of traffic crashes, including those

published in The CMAJ.(3-5)

For Peer Review Only

Page 3 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

COMMENTS

2

MC costs

3) Data abstraction for diagnosing motor vehicle crashes being intuitively and practically

more straightforward than using chart records to diagnose medical conditions such as

diabetes,(6) hypertension,(7) chronic obstructive pulmonary disease (COPD),(8) or

congestive heart failure (CHF);(9) diagnoses which have been formally validated in Ontario’s administrative data.

4) Ontario hospitals receiving additional payments for road injuries. This includes cases

related to motorcycle crashes. MVC codes are thus tied to reimbursement and carefully

scrutinized accordingly.

5) Relative comparisons of motorcycle and automobile crashes using different (police

report) data are similar to those we found. These include reports by Ontario’s Ministry

of Transportation (i.e. ORSAR)(1) and the National Highway Traffic Safety Administration

(NHTSA) in the United States.(10) According to the NHTSA, MCs in the United States are

associated with 6x the death rate as ACs per registered vehicle.(11) Similarly, according

to the 2013 ORSAR, MCs in the Ontario are associated with 4x the death rate as ACs per

registered vehicle.(1) We found 5x the death rate of MCs versus ACs per vehicle.

Including 'mopeds, motor scooters, and motorized bicycles’ in our definition of ‘motorcycles’ is

discussed in Supplementary Appendix B of our paper. However, due to the page limit of the

Interpretation section referenced in Comment 5 below, and the priority given to other

Limitations already discussed in our manuscript, no additional changes to the manuscript have

been made in response to this comment. However, we can add a note in the main text about

including lower speed motorcycles if the Editors deem it prudent.

Comment 2. Many of your drivers will be young and not seeing doctors at all. What

proportion of the young Ontario population is that true for? Are they different from the

others, that you used for controls?

The rationale for excluding those without any contact within 5 years in the first place was to

exclude false health cards, individuals who died and whose death was not registered, as well as

those who left the province without notifying the Ontario Health Insurance Plan. This is

standard practice in costing research using Ontario’s administrative data.(12)

In addition, though this point is well-taken, it is not what we observed in the study. Healthcare utilization in the year prior to the index date, as measured by health care costs accrued during the year prior, was similar between MC-cases and controls (please see Table 3 in the ‘R1’ version of the manuscript or the updated Supplementary Appendix E in updated ‘R2’

manuscript; mean (SD) costs being 1779.76 (7003.45) and 1664.34 (8732.5), standardized

difference = 0.02). The difference-in-differences / self-matching approach would have further

removed this difference in utilization, had it existed, after propensity-score matching.

Comment 3. It appears from your data that a much higher proportion of motorcycle drivers

than automobile drivers will have died on the scene. If this is correct please comment on the

effect of this death rate difference on your cost findings.

For Peer Review Only

Page 4 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

COMMENTS

3

MC costs

As indicated in the Limitations section of the manuscript: ‘We identified patient encounters

with health care providers. As a result, our data does not include those injured and/or killed at

the scene and not transported to hospital.’ Nonetheless, by comparing our data to police report

data from MTO reports (ORSARs) published during the study period, we may infer proportion of

deaths occurring prehospital.(1) We agree this may create an interesting competing outcome,

with rate of early death (without any cost) being higher in motorcycle group. However, this

would underestimate costs in the motorcycle group and we argue is not critical to our payer

perspective analysis (though would be if the perspective were changed from payer to society).

Most importantly, when we actually compare our data to the ORSAR data during the study

period, we calculate 40% of motorcycle deaths (at 30 days) occurred ‘prehospital’, as compared

to 50% automobile deaths (at 30 days).

Comment 4. Please use the STROBE checklist for the revision.

We have uploaded a revised version of the STROBE checklist to reflect the most recent

manuscript version.

Comment 5. Interpretation. This section is about twice as long as we have space for. Can you

please shorten it to as near 2 pages as possible, particularly the Implications section.

The

Interpretation

section

has

been

shortened

to

800

words.

Comment 6. Visual Instruments. We will have room for a maximum of 4 in the published

paper. I suggest you join Tables 1a and 1b. Table 3 could feature in the appendices. Figure 2

could also go to the appendices. Please tell us what you think about these suggestions.

We agree. Thank you for these suggestions. We have edited the updated manuscript,

figures/tables,

changes.

and

supplementary

appendix

reflecting

these

Comment 7. Tables. Please remove shading and bolding from all tables. In Table 2 could you

please supply the N for each column?

These

changes

have

been

made.

Comment 8. Figures. Please make sure that each is supplied in an editable format.

-- Figure 1: please confirm it's original to you, and unpublished elsewhere. We will need an

editable Word or PowerPoint version

We confirm the Figure is original and we have provided it in an editable PowerPoint file.

-- Figure 2: please supply as an editable Excel file with a retained link to the data

An excel file containing this data for this Figure (now Supplementary Appendix E) is included.

For Peer Review Only

Page 5 of 64

COMMENTS

4

MC costs

1

2

3

 

--

4

5

6

Reviewer 1:

7

8

Comment 1. p. 5: it looks like you did both propensity matching for some variables, and

standard variables matching for others (age, sex). Does Rosenbaum/Rubin’s theorem hold

decimal places you matched for, or at least the algorithm you used, and its relative

9

10

11

for such mixed matching? (perhaps you could prove that in the next revision). You need to

12

be more explicit about what the propensity is here, that you are matching on. For MC, it

13

14

15

should be the likelihood of a MC accident (give the degree of propensity scores in terms of

16

success). For AC, it should be the likelihood of a AC accident. (Can you be anymore sure that

17

your controls had no MC or ACs in the sample period? Or does patient confidentiality

18

preclude

matching

with

local

police

citations?)

19

20

21

Greedy matching occurred 1:1 on age (+/- 90 days), sex, and the logit of a propensity score with

22

a caliper of 0.2*standard deviation.(13, 14) Propensity scores were calculated based on patient

23

24

comorbidity, income quintile, and residential location. We have updated the manuscript to

25

reflect these details.

 

26

27

Success of our matching approach in eliminating known confounding was assessed by

examining the balance of covariates after matching.(15-17) Please see our response to

28

29

30

Comment 8 below for a more detailed discussion of the method we used to assess whether

31

covariates were balanced after matching (i.e. standardized differences). As we discuss below,

32

the use of standardized differences to assess the success of / balance of prognosis after

matching is standard practice(16) and has been used in prior studies recently published in The

33

34

35

CMAJ.(18, 19) Several studies indicate factors related to the outcome, rather than the

36

exposure, may improve the balancing ability of the propensity score.(20, 21) Several studies

37

have also found that the discriminatory power of propensity score models such as those

measured by c-statistics, provides no information about whether the model has been correctly

38

39

40

specified, as long as balance of covariates is achieved.(13) Another measure of relative success

41

we discuss in the manuscript is that 98% of cases were matched to controls.

 

42

43

44

Balance of observed covariates does not necessarily mean unobserved covariates were also

standardized difference = 0.02).

45

balanced. However, the difference-in-differences (or self-matching) approach would have

46

further removed individual variation and unmeasured confounding between cases and controls,

47

48

if any existed, after propensity-score matching. However, this was unlikely the case as

49

healthcare utilization in the year prior to the index date (measured by health care costs accrued

50

during the year prio), was similar between MC-cases and controls (please see Table 3 in the ‘R1’

51

version of the manuscript or the updated Supplementary Appendix E in updated ‘R2’

52

53

54

manuscript; mean (SD) costs being 1779.76 (7003.45) and 1664.34 (8732.5) in the year prior,

55

56

Regarding the absence of MCs or ACs among controls during the study period, we can be sure

57

58

controls did not have any health interactions resulting from these collisions because Ontario

59

60

For Peer Review Only

Page 6 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

COMMENTS

5

MC costs

residents have all their medically necessary health care services, provider information, and

demographic characteristics recorded in the health administrative databases used in this study.

As indicated in the Limitations section of the manuscript, however, we only identified patient

encounters with health care providers, and as a result, our data does not include those injured

who did not present to hospital. We do not have access to police reports as suggested.

However, persons involved in MVCs but not presenting to hospital by definition would have not

incurred publicly funded medical costs and not have affected our cost estimates.

Comment 2. p. 5-6 you are fond of citing other “validated” measures used in your study,

without reference to validation in your own study (references 23-31, roughly). Could you add

an index indicating how your sample was similarly validated, i.e., that there is external

validation (between studies) in the use of these indices, in an additional appendix. Thanks.

Since we are not aware of similar studies of health care interactions of MCs and ACs from a

population based sample, there may be differences between our findings and those obtained

from police report data [such as those published Ontario’s Ministry of Transportation (MTO)(1)

and the National Highway Traffic Safety Administration (NHTSA) in the United States].(10)

These differences are due to the different type of data we collected in this study compared to

data collected by these organizations. We identified patient encounters with health care

providers whereas the MTO / NHTSA uses police reports. Our data does not include those killed

at the scene and not transported to hospital and MTO / NHTSA data omit patients injured from

MVCs when police reports are not filed.

Nonetheless, though these reports do not include costs, our findings regarding the additional

risk of MCs are similar to those reported by the MTO and NHTSA. According to the NHTSA, MCs

in the United States are associated with 6x the death rate as ACs per registered vehicle (see

Table 2 from the most recent Motorcycle Traffic Safety Fact Sheet). (11) Similarly, according to

the 2013 ORSAR, MCs in the Ontario are associated with 4x the death rate as ACs per registered

vehicle.(1) We found 5x the death rate of MCs versus ACs per vehicle.

Comment 3. p. 6, and study limitations: you indicate that you would like to capture more (p 10) ongoing costs associate with these injuries; but couldn’t you get a partial handle on this

by comparing the robustness of your results with and without rehab costs, and LTC costs (p

24) that you have data on now. If those weren’t greatly different, than other long term costs

may also not be greatly different between MC and AC; or vice versa.

Although continuing care costs were not significantly different between MCs and ACs from a clinical perspective (see Table 3), inpatient rehab costs (which are reported as part of “acute care hospital” costs in the manuscript) were more than twice as great for MCs compared to ACs within 1-year [$670 (95% CI, 505 – 836) versus $295 (95% CI, 149 – 441), p = <0.001]. The

majority of brain rehabilitation over the long-term in our province is provided on an outpatient

basis and there is still this limitation, which may be differential between MCs and ACs. As we

discuss in the manuscript: “the most important limitation of our cost calculations was our

inability to capture on-going care costs of patients requiring outpatient rehabilitation, such as

For Peer Review Only

Page 7 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

COMMENTS

6

MC costs

those suffering acquired brain injury (ABI).” However, as we also discuss: “Since MC victims are

significantly more likely to suffer head trauma,(22-24) and severe injuries in general (Table 2),

this limitation only underestimated MC-attributable costs compared to those incurred by ACs.”

Comment 6. p. 5, two year followup. Does it matter if its longer? Do you have more data to

check?

As above.

Comment 7. p. 7: its not accidents per vehicle, but accidents per mile driven that are

relevant. One way to get a robustness check on this, is compare seasonal variation. A lot

less M miles driven in the winter PER motorcycle, then A miles driven per auto.

This is an interesting point. Using the most recently published NHTSA data from 2015, the

average registered automobile annually travels approximately 5x the distance compared the

average registered motorcycle.(11) All rates expressed per registered vehicle from our study

could thus be multiplied 5x to be expressed per kilometer (or miles) travelled.

In response to this comment we have added the following point to the discussion section of the

updated manuscript: “All rates in our study could also be multiplied by 5 and expressed per

kilometer travelled since the average automobile travels 5 times the distance of the average

motorcycle (i.e. 15x injuries, 50x severe injuries, 25x deaths, 30x costs per kilometer

travelled).(11)”

Comment 8. P 8 and table 3 are not standard comparisions: you should use the classical

Neymann statistics here and use t-statistics to compare the means. If you think that is too

easy, you then need to bootstrap the standard errors around your SD measures (I presume

this is for standard deviations), and show that those standard deviation measures are not

different from zero. (they seem large by my experience with standardized betas in health

research…)

care

Please note this table is now included as Supplementary Appendix E in the updated manuscript as per the Editors’ suggestion. As discussed in the Methods section of the manuscript: “Baseline characteristics of cases and controls were reported as means and proportions and compared by using standardized differences (greater than 0.1 being considered indicative of imbalance).(33, 34)” Standardized differences (also known as ‘standardized mean differences’ or ‘standardized differences of the mean’) are calculated as the mean difference (mean group 1 – mean group 2) divided by the standard deviation of the measurements. Standardized differences have been shown to provide a measure of effect size and are thought to be particularly useful in large cohort studies using administrative data to avoid identifying clinically spurious statistical

associations because the statistic is less influenced by sample size compared to statistical

significance/classical Neymann statistics.(16) With statistical significance/classical Neymann

statistics, in contrast, precision increases and standard errors decrease with increases in sample

size and no measure of effect size is provided.(16) The use of standardized differences to

For Peer Review Only

Page 8 of 64

COMMENTS

7

MC costs

1

2

3

assess balance of prognosis after matching is standard practice,(16) including prior studies

4

5

recently published in The CMAJ.(18, 19)

6

7

Comment 9. Also, explain to the reader why you have just a one to one match here; many to

one matches are more efficient for estimation (though 1 to 1 is asymptotically valid, I

8

9

10

agree). Note in the quintiles that your controls are poorer for both groups, but more

11

especially for the MC group. Could you speculate why this is, and how it might bias the

12

comparison (since it’s a differential response between AC and MC)?

13

14

15

We chose 1:1 matching to facilitate comparisons of covariates and easily and intuitively

16

evaluate the success of our matching algorithm.(15-17) Although we agree adding controls may

17

even further increase the precision of our standard errors, our study was already adequately

powered using a 1:1 match to detect statistically significant differences between groups.

18

19

20

Indeed, one reason we used standardized differences for comparisons in our study was to avoid

21

identifying clinically spurious statistical associations in the large dataset.(16)

22

23

24

Income inequality between cases and controls is an important point. We do acknowledge that

25

although most baseline characteristics were balanced between cases and controls after

26

matching, MC cases were more likely to live in higher income neighborhoods. The clinical

27

relevance of 20.5% cases versus 16.4% controls living in the highest income neighborhoods,

however, is difficult to contextualize. The fact that motorcycle insurance is already rising in our

28

29

30

province may explain part of this slight imbalance.(25) Low SES has been shown to predict

31

worse outcomes and higher healthcare utilization in our province.(26-28) As such, poorer MC-

32

controls likely would have underestimated MC-attributable costs if anything.

33

34

35

Nonetheless, and as we mentioned above, despite statistically significant differences in income

36

between MC-cases and controls, healthcare utilization in the year prior to the index date was

37

similar between MC-cases and controls (please see Table 3 in the ‘R1’ version of the manuscript

or the updated Supplementary Appendix E in updated ‘R2’ manuscript; mean (SD) costs being

38

39

40

1779.76 (7003.45) and 1664.34 (8732.5), standardized difference = 0.02). The difference-in-

41

differences / self-matching approach would have further removed this difference in utilization,

42

43

had it existed, after propensity-score matching.

44

45

Comment 10. Table 2: why abbreviate in the heading when there is plenty of room to spell

46

out ISS and AIS, rather than add those in the footnote here.

47

48

49

These changes were made.

50

51

--

52

53

54

55

Reviewer 2:

56

57

58

59

60

For Peer Review Only

Page 9 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

COMMENTS

8

MC costs

Comment 1. a) The calculation of cost differences at the top of page 6 and in figure 1 were to

take into account any cost inflation differences between the pre year and post-two study

years.

We apologize this was not clear in the initial version of the manuscript but all costs were

assigned to each patient based on the year when they were incurred and then inflated to 2013

Canadian dollars using the health care component of the Ontario consumer price index (CPI,

www.statscan.gc.ca). In response to this comment, the following has been added to the

updated methods section: “All costs were expressed as 2013 Canadian dollars using the health

care component of the Ontario consumer price index (CPI, www.statscan.gc.ca).“

Comment 2. b) The authors would add a small calculation as to the costs, mortality rates, serious injury rates PER 100,000 MC or AC vehicle kilometers travelled. I am sure data is

available as to the average annual kilometrage of both modes.

This is an interesting point. Using the most recently NHTSA data from 2015, the average

registered automobile travels approximately 5x the distance annually compared the average

registered motorcycle.(11) All rates expressed per registered vehicle in our study could thus be

multiplied 5x to be expressed per kilometer (or miles) travelled.

In response to this comment we have added the following point to the discussion section of the

updated manuscript: “Since the average automobile annually travels 5 times the distance of the

average motorcycle, all rates in our study could also be multiplied by 5 and expressed per

kilometer travelled (i.e. 15x injuries, 50x severe injuries, 25x deaths, 30x costs per kilometer

travelled).(11)”

Comment 3. Middle of page 9:

write "were double" instead of "were 2 times greater"

This change (as well as “3 times greater” -> “triple“) has been made to the updated manuscript.

Comment 4. Abstract: last time : ADD

motorcycle safety.

This change was made.

--

Reviewer 3:

may provide AN ADDITIONAL incentive to improve

Introduction Comment 1. - (Major) Similar studies and background literature do not appear to be

introduced and referenced completely. Though studies on motorcycle crash costs may be

limited, it would be valuable to discuss what is known about prevalence and cost of both MCs

and ACs. A preliminary Medline search resulted in several relevant papers on MCs alone that

For Peer Review Only

Page 10 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

COMMENTS

9

MC costs

have not been cited. There are similar cost analyses for automobile crashes as well that

should be introduced. For the articles cited, there is lack of detail on what these studies tell

us.

We have cited in the introduction of our manuscript the systematic review of prior studies

examining medical costs of motorcycle crashes conducted by the National Highway Traffic

Safety Administration (NHTSA) in 2003. This systematic review concluded that prior calculations

of medical costs attributable to MCs are limited to reviews of hospital charges at single centers

and that estimates derived from these reviews were incomplete, neglecting costs incurred after

a patient’s discharge.(24, 29, 30) In other words, that no reliable estimate of medical costs

incurred by MCs exists.(30) We have been unable to identify any research on this topic

published since that time. There is also no literature of which we are aware that directly

compares costs of MCs and ACs.

In our study, we examined all health care costs accrued during each patient’s index

visit/admission, any subsequent readmissions/visits, costs for continuing care including

residential long-term care and home care, and costs for rehabilitation and physicians, all being

assessed up to 2 years after the injury. Our study was also population based in contrast to the

prior studies on this topic which recorded the hospital charges of inpatients at single centers,

primarily trauma centers.(30) As a result, we were able to study the vast majority of MC and AC

cases who are treated as outpatients and do not have severe injuries (>90%) that were

excluded in prior studies.

In contrast to data on costs, other reports of clinical data regarding the additional risk of MCs

are publically available. According to the NHTSA, MCs in the United States are associated with

6x the death rate as ACs per registered vehicle (see Table 2 from the most recent Motorcycle

Traffic Safety Fact Sheet). (11) Similarly, according to the 2013 ORSAR, MCs in the Ontario are

associated with 4x the death rate as ACs per registered vehicle.(1) We found 5x the death rate

of MCs versus ACs per vehicle. However, these other reports do not consider costs or provide

detailed descriptions of injury types (using the Abbreviated Injury Scale, for example).

We have tried to frame our study in the updated Introduction / Interpretation sections within

the existing literature(30) and publically available data on this topic.(1, 10) We do not think

including details of this literature (of which we are aware) beyond what is presented above and

in the updated manuscript is prudent given the word limit of the manuscript.

Methods Comment 2. - Page 4 lines 31-36: Is there any literature to show that ICD-10 codes related to motorcycle and automobile crashes are assigned accurately? It would be helpful to know if these codes have been validated in existing literature, and if so, what the positive predictive

value

is.

This is an important point. Please see our response to the Editorial Board’s Comment 1 above.

For Peer Review Only

Page 11 of 64

1

2

3

4

5

COMMENTS

10

MC costs

Comment 3. - Page 4 line 44: It would be valuable to clarify what you mean by “30% general

6

sample”.

This

may

not

come

across

intuitively

to

readers.

7

8

9

The

following change has been made to the manuscript in response to this comment: “For each

10

MC

and AC case, we identified one control from a representative subset of the Ontario

11

population.”

 

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

Comment 4. - Page 4 line 51: The control group consisted of patients who had at least some

health system contact during the appropriate inclusion years. This control group may not be

fully representative of the underlying population at risk…. I assume that many otherwise-

healthy men in their 30’s who ride motorcycles, or “average” MC patients, do not have

regular health system contact. How did you justify using this sample? Limitations of this

control group should be addressed in the appropriate section.

This is also an important point. Please see our response to the Editorial Board’s Comment 2

above.

Comment 5. - (Major) Page 5 Lines 3-9: It is unclear how you used the propensity score to

conduct your match. Clarification on this is essential. If using propensity score matching, what

technique of matching was used – ie. nearest matching, greedy nearest neighbour matching

with

etc.

calipers

+/-

replacement,

Greedy matching occurred 1:1 on age (+/- 90 days), sex, and the logit of a propensity score with

a caliper of 0.2*standard deviation(13, 14) and was calculated based on patient comorbidity,

income quintile, and residential location.

We have updated the manuscript to reflect these details.

Comment 6. - Page 5 Line 51: Since this is not a typically-used method, a reference to the

valuable.

“difference-in-differences”

approach

would

be

In response to this comment, we have cited a recent article from JAMA that summarizes the difference-in-differences approach.(31) Although the article focuses on the application of this approach to evaluating the implementation of health policy, the explanations and illustration provided also apply to our study of costs.

Comment 7. - Page 6 Line 18: “All costs were expressed as 2013 Canadian dollars.” Please

reference

rates.

the

source

used

for

inflation

We apologize this was not clear in the original manuscript. Costs were assigned to each patient

based on the year when they were incurred and then inflated to 2013 Canadian dollars using

the health care component of the Ontario consumer price index (CPI, www.statscan.gc.ca). In

response to this comment, the following has been added to the updated methods section: “All

For Peer Review Only

Page 12 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

COMMENTS

11

MC costs

costs were expressed as 2013 Canadian dollars using the health care component of the Ontario

consumer price index (CPI, www.statscan.gc.ca).“

Comment 8. - Page 6 Lines 54-57: Please reference source of proposed 0.1 as indicative of

imbalance.

paper.

I

believe

this

is

from

Normand

et

al.’s

2001

Thank you. This reference has been added to the updated manuscript.(32)

Comment 9. - It would be helpful to specify the source used to determine the number of

registered automobiles and motorcycles (reference 18) in the Methods section.

This was the number of each vehicle registered each year during the study period with

Ontario’s Ministry of Transportation. As mentioned in the manuscript: “Annual rates of injuries

and deaths were expressed per 100,000 registered motorcycles or automobiles in the Province

by using publicly available data from the Ministry of Transportation (MTO) (see Supplementary

Appendix C).(1)”

Results

Comment 10. - Page 8 lines 26-28: “Covariates were balanced between cases and controls,

the exception being that MC cases were more likely to reside in rural areas than controls.”

Please expand on this and potential implications, if any, in Limitations section. Were any

other

done?

model

diagnostics

This is an important point. Although most baseline characteristics were balanced between cases

and controls after matching, MC cases were significantly more likely to reside in rural areas

than controls. However, the clinical relevance of this difference between a rurality index of 15.9

versus 11.3 is difficult to contextualize since a rurality index >40 is generally considered

indicative of ‘rural residence’.(33) This finding may be the result of motorcycles being difficult

to ride in the city (and therefore) being more common in the country. Nonetheless and as

discussed above, healthcare utilization in the year prior to the index date, as measured by health care costs accrued during that year, was actually similar between MC-cases and controls (please see Table 3 in the ‘R1’ version of the manuscript or the updated Supplementary Appendix E in updated ‘R2’ manuscript; mean (SD) costs being 1779.76 (7003.45) and 1664.34 (8732.5), standardized difference = 0.02). The difference-in-differences / self-matching approach would have further removed this difference in utilization, had it existed, after propensity-score matching.

Regarding model diagnostics, several studies have found that the discriminatory power of propensity score models such as those measured by c-statistics, provides no information about whether the model has been correctly specified, as long as balance of covariates is

achieved.(13) Therefore, the success of our matching approach was assessed by examining the

balance of covariates after matching.(15-17) We also report the proportion of eligible cases

matched to controls (approximately 98%).

For Peer Review Only

Page 13 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

COMMENTS

12

MC costs

Due to the page limit of the Interpretation section referenced in the Editorial Board’s Comment

5 above and the priority given to other Limitations already discussed, no changes to the

manuscript have been made in response to this comment.

Interpretation

Comment 11. - (Major) No reference is made to existing literature on MC and AC prevalence

and costs. Though I appreciate that high-quality studies on MC costs may be lacking, relevant

literature based out of Canada and the United States exists and should be introduced and

compared to these findings. Furthermore, many studies have been done internationally (Iran,

Nigeria, Brazil, Taiwan, Mexico, etc.) that would add to the discussion portion. Discussing

how your estimated AC costs relate to findings in current literature would also be valuable.

Please see our response to Comment 1 above.

Comment 12. - Page 11 lines 21-26: “Expressed in other terms, Ontario’s healthcare payer

(MOHLTC) would expect to save $13.5 million dollars per year if the Province’s registered

motorcycles were replaced with automobiles.” This is assuming the rate of automobile

accidents would stay the same. What if “higher-risk” motorcycle drivers are also higher-risk

automobile drivers, and their probability of crashing is higher than the average automobile

driver?

‘Table 6.2: Selected Factors Relevant to Fatal Motorcycle Collisions’ of the 2013 Ontario Road

Safety Annual Report indicates that most motorcycle-related deaths in Ontario are unlikely the

result of modifiable risk factors or risk-taking behaviors, other than choosing to ride a

motorcycle itself. For example, we know that 97.9% of motorcyclists in 2013 were wearing

helmets at the time of their crash.(1)

Based on this fact, and that province-level cost estimates may provide context for readers and

policymakers, we have kept this discussion point. However, our assumption of a constant

accident rate is a good point and we have updated the manuscript to reflect this: “Expressed in

other terms, Ontario’s healthcare payer (MOHLTC) would expect to save $13.5 million dollars per year if collision rates remained unchanged and the Province’s registered motorcycles were

replaced

automobiles.”

with

Comment 13. - Page 12 lines 24-29: How do you anticipate that a tax would increase safety? It may deter drivers with a lower income, but I cannot appreciate how it would inherently

safety.”

“improve

motorcycle

As suggested, an excise tax may deter motorcycle use, similar to the way tobacco prices have been one of the most effective means of reducing tobacco use.(34-36) Anecdotally, raising

insurance rates in response to private medical costs has already acted as a deterrent to

motorcycle use in our jurisdiction.(25) There are other potential policy initiatives we discuss in

the manuscript, such as supplementary private health insurance and costs recoverable from

For Peer Review Only

Page 14 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

COMMENTS

13

MC costs

motorcycle insurers under the Province’s Insurance Act. The broader expression ‘improving

motorcycle safety’ has been used instead of deterrence for this reason.

Comment 14. - You did not match based on previous number/type of health care encounters.

You may want to add in your discussion how costs in the previous year compare for cases

versus controls. This may help strengthen the argument that you have comparable groups.

Thank you for this thoughtful point. We have added the following additional analysis to the

manuscript in response: “Baseline healthcare utilization as measured by costs in the year prior

was similar between cases and controls, also indicating groups were comparable (mean (SD)

prior year costs for MC-cases and controls were 1779.76 (7003.45) and 1664.34 (8732.5),

standardized difference = 0.02; ACs-cases and controls 2385.69 (8223.62) and 2314.57

(9420.75) standardized difference = 0.01).”

As mentioned above, the difference-in-differences / self-matching approach was intended to

remove residual differences in health care utilization between cases and controls, had they

existed, after propensity-score matching.

Comment

- Table 2: Does the severe injuries score take into account injuries noted during the first

assessment or the entire health care visit/admission? In other words, are initially-missed

for?

Tables

15.

injuries

accounted

This is an important point. Any injury diagnoses noted during the patients emergency

department visit (in the NACRS database) and inpatient stay (in the CIHI-DAD database) were

accounted for using an algorithm previously validated in Ontario’s administrative data.(37)

Initially-missed injuries were thus accounted for. Please see Supplementary Appendix A for

more information about the different administrative databases used in this study.

Comment 16. - Table 3: The standard deviations for average cost appear quite large. Are these accurate? If so, we lose the ability to calculate costs at the lower end of the spectrum. 95% confidence intervals would show the range better.

The updated Table 3 now contains 95% confidence intervals in response to this suggestion. We agree the standard deviations for average costs are large, but these are accurate and similar to cost ranges previously published for different patient populations using the same methodology.(38) The distribution is likely due to the wide range of pathology (and associated costs) sampled in our study, ranging from those requiring only an ED visits to those admitted to the ICU and requiring inpatient rehabilitation. Similarly, these severe injuries that required admission to hospital and the ICU likely accounted for the additional costs incurred by MCs

compared to ACs.

For Peer Review Only

Page 15 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

COMMENTS

14

References for Comments

MC costs

1. Ministry of Transportation. Ontario Road Safety. Annual Report 2013.

2. Juurlink D, Preyra C, Croxford R, Chong A, Austin P, Tu J, Laupacis A. Canadian Institute

for Health Information Discharge Abstract Database: A Validation Study. Toronto: Institute for

Clinical Evaluative Sciences; 2006.

3. Redelmeier DA, Katz D, Lu H, Saposnik G. Roadway crash risks in recent immigrants.

Accid Anal Prev. 2011;43(6):2128-33.

4. Redelmeier DA, May SC, Thiruchelvam D, Barrett JF. Pregnancy and the risk of a traffic

crash. CMAJ. 2014;186(10):742-50.

5. Bai YQ, Santos G, WP W. Cost of Public Health Services for Ontario Residents. Applied

Health Research Question Series. Toronto: Health System Performance Research Network;

2016.

6.

Hux JE, Ivis F, Flintoft V, Bica A. Diabetes in Ontario: determination of prevalence and

incidence using a validated administrative data algorithm. Diabetes care. 2002;25(3):512.

7. Tu K, Campbell NR, Chen ZL, Cauch-Dudek KJ, McAlister FA. Accuracy of administrative

databases in identifying patients with hypertension. Open medicine : a peer-reviewed,

independent, open-access journal. 2007;1(1):e18.

8. Gershon AS, Wang C, Guan J, Vasilevska-Ristovska J, Cicutto L, To T. Identifying

individuals with physcian diagnosed COPD in health administrative databases. Copd.

2009;6(5):388.

9. Ko DT, Mamdani M, Alter DA. Lipid-lowering therapy with statins in high-risk elderly

patients: the treatment-risk paradox. Jama. 2004;291(15):1864.

10. The National Highway Traffic Safety Administration. TRAFFIC SAFETY FACTS 2015.

Available from https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/812384. Accessed

April 27, 2017.

11. NHTSA’s National Center for Statistics and Analysis (2015), NHTSA: Motorcycles Traffic

Safety Fact Sheet , Accessed June 24, 2017

from https://crashstats.nhtsa.dot.gov/Api/Public/ViewPublication/812353.

12. Wodchis WP, Austin PC, Henry DA. A 3-year study of high-cost users of health care.

CMAJ. 2016;188(3):182-8.

13. Austin PC. An Introduction to Propensity Score Methods for Reducing the Effects of

Confounding in Observational Studies. Multivariate behavioral research. 2011;46(3):399.

14. Austin PC. Optimal caliper widths for propensity-score matching when estimating

differences in means and differences in proportions in observational studies. Pharmaceutical

statistics. 2011;10(2):150.

15. Austin PC. Using the standardized difference to compare the prevalence of a binary

variable between two groups in observational research. Communications in Statistics:

Simulation and Computation. 2009;38(6):1228.

16. Austin PC. Balance diagnostics for comparing the distribution of baseline covariates

between treatment groups in propensity-score matched samples. Statistics in medicine.

2009;28(25):3083.

17. Fitzmaurice G. Confounding: Propensity score adjustment. Nutrition. 2006;22(11-

12):1214.

For Peer Review Only

Page 16 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

COMMENTS

15

MC costs

18. Taipale H, Tolppanen AM, Koponen M, Tanskanen A, Lavikainen P, Sund R, et al. Risk of

pneumonia associated with incident benzodiazepine use among community-dwelling adults

with Alzheimer disease. CMAJ. 2017;189(14):E519-E29.

19. Shih CJ, Chen HT, Kuo SC, Li SY, Lai PH, Chen SC, et al. Comparative effectiveness of

angiotensin-converting-enzyme inhibitors and angiotensin II receptor blockers in patients with

type 2 diabetes and retinopathy. CMAJ. 2016;188(8):E148-57.

20. Brookhart MA, Schneeweiss S, Rothman KJ, Glynn RJ, Avorn J, Stürmer T. Variable

selection for propensity score models. American Journal of Epidemiology. 2006;163(12):1149.

21. Rubin DB. On principles for modeling propensity scores in medical research.

Pharmacoepidemiology and drug safety. 2004;13(12):855.

22. Keng SH. Helmet use and motorcycle fatalities in Taiwan. Accident Analysis and

Prevention. 2005;37(2):349.

23. Hotz GA, Cohn SM, Mishkin D, Castelblanco A, Li P, Popkin C, et al. Outcome of

Motorcycle Riders at One Year Post-Injury. Traffic Injury Prevention. 2004;5(1):87.

24. Monk JP, Buckley R, Dyer D. Motorcycle-related trauma in Alberta: A sad and expensive

story. Canadian Journal of Surgery. 2009;52(6):E235.

25. Harvey I. Motorcycle rates rising with severity, number of claims: Toronto Star; 2016

[Available from: https://www.thestar.com/autos/2016/05/28/motorcycle-rates-rising-with-

severity-number-of-claims.html.

26. Agabiti N, Picciotto S, Cesaroni G, Bisanti L, Forastiere F, Onorati R, et al. The influence

of socioeconomic status on utilization and outcomes of elective total hip replacement: a

multicity population-based longitudinal study. International journal for quality in health care :

journal of the International Society for Quality in Health Care / ISQua. 2007;19(1):37.

27. Santaguida PL, Hawker GA, Hudak PL, Glazier R, Mahomed NN, Kreder HJ, et al. Patient

characteristics affecting the prognosis of total hip and knee joint arthroplasty: a systematic

review. Canadian journal of surgeryJournal canadien de chirurgie. 2008;51(6):428.

28. Matheson FI, Moineddin R, Dunn JR, Creatore MI, Gozdyra P, Glazier RH. Urban

neighborhoods, chronic stress, gender and depression. Social science &amp; medicine (1982).

2006;63(10):2604.

29. Dobson JL. THE PUBLIC COST OF MOTORCYCLE TRAUMA. Jama-Journal of the American

Medical Association. 1989;261(8):1149-.

30. Administration NHTS. Costs of Injuries Resulting from Motorcycle Crashes: A Literature

Review 2003 [Available from:

http://www.nhtsa.gov/people/injury/pedbimot/motorcycle/Motorcycle_HTML/overview.html.

31. Dimick JB, Ryan AM. Methods for evaluating changes in health care policy: the

difference-in-differences approach. JAMA. 2014;312(22):2401-2.

32. Normand ST, Landrum MB, Guadagnoli E, Ayanian JZ, Ryan TJ, Cleary PD, et al. Validating

recommendations for coronary angiography following acute myocardial infarction in the

elderly: a matched analysis using propensity scores. J Clin Epidemiol. 2001;54(4):387-98.

33. Kralj B. Measuring “Rurality” for Purposes of Health Care Planning: An Empirical

Measure for Ontario. Toronto: Ontario Medical Association. 2005.

34. Chaloupka F. Rational Addictive Behavior and Cigarette-Smoking. Journal of Political

Economy. 1991;99(4):722-42.

For Peer Review Only

Page 17 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

COMMENTS

16

MC costs

35. Gilmore AB, Tavakoly B, Taylor G, Reed H. Understanding tobacco industry pricing

strategy and whether it undermines tobacco tax policy: The example of the UK cigarette

market. Addiction. 2013;108(7):1317-26.

36. Jha P, Chaloupka FJ. The economics of global tobacco control. British Medical Journal.

2000;321(7257):358-61.

37. Haas B, Xiong W, Brennan-Barnes M, Gomez D, Nathens AB. Overcoming barriers to

population-based injury research: Development and validation of an ICD-10-to-AIS algorithm.

Canadian Journal of Surgery. 2012;55(1):21.

38. Rosella LC, Fitzpatrick T, Wodchis WP, Calzavara A, Manson H, Goel V. High-cost health

care users in Ontario, Canada: demographic, socio-economic, and health status characteristics.

BMC health services research. 2014;14:532.

For Peer Review Only

Page 18 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

MANUSCRIPT

1

MC costs

Direct Medical Costs of Motorcycle Crashes in Ontario

Authors:

Daniel Pincus 1,2,3 MD

David Wasserstein 1,4 MD, MSc, MPH

Avery B. Nathens 1,2,3,4 MD, MPH, PhD

Yu Qing Bai 2,3 MSc

Donald A. Redelmeier 2,3,5,6 MD, MS(HSR)

Walter P. Wodchis 2,3 PhD

Affiliations:

1 Department of Surgery, University of Toronto, Toronto, Canada 2 Institute for Clinical Evaluative Sciences, Toronto, Canada

3 Institute of Health Policy, Management and Evaluation, University of Toronto, Toronto, Canada

4 Department of Surgery, Sunnybrook Health Sciences Centre, Toronto, Canada

5 Division of General Internal Medicine, Sunnybrook Health Sciences Centre, Toronto, Canada,

6 Center for Leading Injury Prevention Practice Education & Research, Toronto, Canada

Correspondence and address for all authors:

Daniel Pincus MD

Department of Surgery, Division of Orthopaedic Surgery, University of Toronto

149 College Street, Room 508-A

Toronto, ON M5T 1P5

Phone: 416-946-7957 (w), 647-244-3324 (c)

Fax: 416-978-3928

d.pincus@utoronto.ca

Competing Interests / Disclosures:

There are no relevant disclosures or competing interests. All authors meet ICMJE criteria.

Source of funding:

This research was supported by grants from the Ontario Ministry of Health and Long-Term Care (MOHLTC) and the Ontario SPOR Support Unit to the Health System Performance Research Network (HSPRN: fund #06034, recipient WPW), and by the Institute for Clinical Evaluative Sciences (ICES), which is also funded by an annual grant from the MOHLTC. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Parts of the material are based on data and information compiled and provided by the Canadian Institute for Health Information (CIHI). No endorsement by ICES, the MOHLTC or CIHI is intended or should be inferred. No benefits have been received or will be received from a commercial party related directly or indirectly to the subject of this article.

Key Words: Motorcycles, Accidents, Traffic/economics, Traffic/statistics & numerical data,

Insurance, Health/economics, Retrospective Studies

For Peer Review Only

Page 19 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

MANUSCRIPT

2

ABSTRACT (Word Count: 246)

MC costs

Background. No reliable estimate of costs incurred by motorcycle crashes (MCs) exists. Our

objective was to calculate the direct costs of all publicly funded medical care provided to

individuals following MCs compared to automobile crashes (ACs).

Methods. We conducted a population based, matched cohort study of adults in Ontario that

presented to hospital because of a MC or AC from 2007 through 2013. For each case, we

identified one control absent a motor vehicle crash during the study period. Direct costs for

each case and control were estimated as 2013 $ Canadian from the payer perspective using

methodology that links healthcare use to individuals over time. MC- and AC-attributable costs

within 2 years were then calculated using a difference-in-differences approach.

Results. We identified 26,831 patients injured from MCs and 281,826 injured from ACs. Mean

MC- and AC-attributable costs were $5,825 and $2,995, respectively (p<0.0001). The rate of

injury was triple for MCs compared to ACs (2,194 injured yearly/100,000 registered motorcycles

versus 718 injured yearly/100,000 registered automobiles; IRR=3.1, 95% CI=2.8-3.3, p<0.0001).

Severe injuries, defined as those with an Abbreviated Injury Scale >=3, were 10 times greater

(125 severe injuries per yearly/100,000 registered motorcycles versus 12 severe injuries per

yearly/100,000 registered automobiles; IRR=10.4, 95% CI=8.3-13.1, p<0.0001).

Interpretation. Considering both the attributable cost and higher rate of injury, we found each

registered motorcycle in Ontario costs the public healthcare system 6 times the amount of each

registered automobile. Medical costs may provide an additional incentive to improve

motorcycle safety.

For Peer Review Only

Page 20 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

MANUSCRIPT

INTRODUCTION

3

MC costs

Motor vehicle safety has improved significantly over the past 100 years.(1) Between

2000 and 2010, for example, mortality related to motor vehicle crashes (MVCs) decreased by

55.1% in 19 developed countries.(1) In contrast, deaths and injuries among a subset of MVC

patients – those injured in motorcycle crashes (MCs) – remained stable during the same time

period.(2)

Although medical costs may provide incentive to improve motorcycle safety,(3-5) no

reliable estimate of medical costs incurred by MCs exists. Prior calculations of medical costs

attributable to MCs are limited to reviews of hospital charges at single centers.(6) Estimates

derived

from

these

reviews

are

incomplete,

discharge, for example.(6-8)

neglecting

costs

incurred

after

a

patient’s

The calculation of patient level medical costs in Ontario, Canada (population 13.6 million

in 2014) is now possible with methodology that links publicly funded healthcare use to

individuals over time.(9) Using this methodology from the payer perspective, our objective was

to calculate the direct costs of all publicly funded medical care provided to individuals

presenting to hospital following MCs in comparison to automobile crashes (ACs). We also

examined the population incidence of injuries resulting from these crashes. Our hypothesis was

that medical costs and injury rates attributable to MCs, which can occur at high speeds with less

personal protection, are significantly higher than those attributable to ACs.

METHODS

Setting

For Peer Review Only

Page 21 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

MANUSCRIPT

4

MC costs

We conducted a population based, matched cohort study in Ontario, Canada using

linked health administrative databases at the Institute for Clinical Evaluative Sciences (ICES)

(see Supplementary Appendix A – ‘data sources’). Ontarians have their medically necessary

health care interactions, provider information, and demographic characteristics recorded in

these databases. These data have previously been used to estimate medical costs(10-15) and

study traffic crashes.(16-18) The research protocol was approved by the Research Ethics Board

at Sunnybrook Health Sciences Centre.

Patients

The study design is displayed in Figure 1. Ontarians who presented to the emergency

department (ED) or were admitted to hospital following a MC or AC between April 1, 2007 to

March 31, 2013 were eligible for inclusion. International Classification of Diseases, 10 th Revision

(ICD-10) codes were used to identify injured motor vehicle occupants [MC (V20-V29) and AC

(V40-V49, V50-V59, V70-V79), respectively].(19) We excluded non-Ontario residents and those

with prior hospital presentation for a motor-vehicle related injury of any type within 2 years

before the index date (see Supplementary Appendix B – ‘database codes’).

For each MC and AC case, we identified one control from a representative subset of the

Ontario population. The purpose of these controls was to determine baseline medical costs

absent a motor vehicle crash and thus enable the calculation of incremental costs. Controls had

health system contact but no documented motor-vehicle related injury during this timeframe

or dating back 2 years prior to the index date. Index dates for controls were randomly assigned

based on the distribution of index dates for cases in the same fiscal year. Greedy matching

For Peer Review Only

Page 22 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

MANUSCRIPT

5

MC costs

occurred on age (+/- 90 days), sex, and the logit of a propensity score with a caliper of

0.2*standard deviation (20-22) and was calculated based on patient comorbidity, income

quintile, and residential location (see ‘Covariates’).

Covariates

We measured several covariates that have been shown to influence health care

utilization in Ontario.(13) Age and sex were analyzed as continuous and categorical variables,

respectively. Comorbidities listed on outpatient and hospital discharge abstracts in the two

years before the index date were categorized according to Collapsed Aggregate Diagnosis

Groups (CADGs).(23) Neighborhood income quintile was used as a validated surrogate measure

for socioeconomic status and social deprivation.(24-26)

Patient location of residence was

classified according to Local Health Integrated Network (LHIN) and the Rurality Index of Ontario

(RIO).(27)

Outcomes

The primary outcome of this study was direct medical costs within 2 years, attributable

to MCs and ACs, and paid by Ontario’s Ministry of Health and Long-term Care (MOHLTC). First,

medical costs were calculated for each case and control during each fiscal year of the study

period using established patient level costing methodology (see Supplementary Appendix A).(9)

MC- and AC-attributable costs within 2 years were then calculated using a difference-in-

differences approach: a baseline cost accrued the year prior to a crash was subtracted from

For Peer Review Only

Page 23 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

MANUSCRIPT

6

MC costs

costs in the first and second year following the crash (first difference) and then compared to

the same difference among controls (second difference).(28)

Costs were categorized as: (a) acute care hospital including the ED, index admission,

rehabilitation, and any subsequent readmissions, (b) physician, (c) drugs/laboratory costs

outside of hospitals, (d) continuing care including residential long-term care and home care,

and (e) assistive devices.(18) Care episodes that spanned more than one fiscal year were

divided on a pro rata basis. All costs were expressed as 2013 Canadian dollars using the health

care component of the Ontario consumer price index (CPI, www.statscan.gc.ca).

Secondary outcomes included clinical data. First, the annual rates of those injured and

dying (within 30 days) from MCs and ACs were reported, so long as the patient presented to

hospital after the crash. More detailed clinical data were also compared between MC and AC

cases. We identified the highest level of care required for each patient: ED visit only,

hospitalization, or intensive care unit (ICU) admission. The Injury Severity Score (ISS) and

Abbreviated Injury Scale (AIS) were used to assess injury severity. ISS and AIS were obtained

from ICD-10 codes by means of a validated algorithm.(29) ISS was categorized as <9, 9-15, 16-

24, or >=25. ‘Severe injuries’ were those with an AIS>=3 and were reported overall and for each

anatomical region. The algorithm and operational definitions have been used previously in

population based research of Ontario’s administrative data.(30-32)

Statistical Analysis

Baseline characteristics of cases and controls were reported as means and proportions

and compared by using standardized differences (greater than 0.1 being considered indicative

For Peer Review Only

Page 24 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

MANUSCRIPT

7

MC costs

of imbalance).(33, 34) We also compared characteristics between MC-cases with AC-cases by

using independent samples t-tests for continuous variables and chi-square tests for categorical

variables. Longer-term costs (up to 5 years) were also compared between matched cases and

controls enrolled before FY 2011 for whom longer-term follow-up (>2 years) was available.

Annual rates of injuries and deaths were expressed per 100,000 registered motorcycles or

automobiles in the Province by using publicly available data from the Ministry of Transportation

(MTO) (see Supplementary Appendix C).(2) Incidence rate ratios (IRRs) with 95% confidence

intervals (CIs) were then used to compare injury and death rates between MC- and AC-cases.

All analyses were performed using SAS software (SAS version 9.3 and SAS Enterprise Guide

version 6.1; SAS Institute) and the type-I error probability was set to 0.05.

RESULTS

Descriptive clinical data; (Table 1, Table 2)

We identified 26,831 Ontarians during the study period injured during MCs and 281,826

injured during ACs. The annual incidence of injured persons was triple for MCs compared to ACs

(2,194 injured per year/100,000 registered motorcycles versus 718 injured per year/100,000

registered automobiles; IRR=3.1, 95% CI=2.8-3.3, p<0.0001; Table 1). Deaths (at 30d) were

approximately 5 times greater (14 deaths per year/100,000 registered motorcycles versus 3

deaths per year/100,000 registered automobiles; IRR=4.7, 95% CI=2.9-7.6, p<0.0001; Table 1).

In terms of the level of care required for each patient, those injured in MCs were

significantly more likely to require hospital and ICU admission than those injured in ACs (11.93%

versus 4.24%, p < 0.0001 and 2.31% versus 1.09%, p < 0.0001, respectively). MC patients were

For Peer Review Only

Page 25 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54

55

56

57

58

59

60

MANUSCRIPT

8

MC costs

also more likely to have an ISS > 16 (3.67% versus 1.32%, p < 0.0001) and suffer severe injuries

(5.69% versus 1.67% with AIS>=3, p < 0.0001). Expressed per registered vehicle, severe injuries

were 10 times greater for MCs than ACs (125 severe injuries per year/100,000 registered

motorcycles versus 12 severe injuries per year/100,000 registered automobiles; IRR=10.4, 95%

CI=8.3-13.1, p<0.0001). Severe head, thorax, abdominal and extremity injuries were all more

common among patients injured in MCs compared to ACs (Table 2).

Baseline characteristics after matching; (Supplementary Appendix E)

There were 26,257 MC cases and 276,760 AC cases (98%) matched to controls.

Covariates were balanced between cases and controls, the exception being that MC cases were

more likely to reside in rural areas than controls. Baseline healthcare utilization as measured by

costs in the year prior was similar between cases and controls, also indicating groups were

comparable (mean (SD) prior year costs for MC-cases and controls were 1779.76 (7003.45)

and 1664.34 (8732.5), standardized difference = 0.02; AC-cases and controls 2385.69 (8223.62)

and 2314.57 (9420.75) standardized difference = 0.01).

The mean age of individuals injured in MCs was younger (mean 36.08) than those

injured in ACs (mean 38.53) (p < 0.0001). A greater proportion of MC patients were also male

(81% male) in comparison to AC patients (43% male) (p < 0.0001). Other baseline characteristics

and average costs amongst cases and controls after matching are presented in Supplementary

Appendix E.

Patient-level healthcare costs; (Table 3, Supplementary Appendix D)

For Peer Review Only

Page 26 of 64

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20