Paramedics’ knowledge, attitude and preparedness to manage out-of-hospital obstetric emergencies in Queensland, Australia: exploratory analysis
Federico Andreis
26/08/2021
Dataset overview
The dataset contains 435 records for 91 variables. The percentage of completed (100% responses) questionnaires is \(\approx\) 64.83%, with \(\approx\) 71.72% completing at least 81% of it. More on missing data in the next section.
Completion time ranged from (rounded values) 0 to 7102 minutes, with a median of 21 minutes (first quartile 11, third quartile 139, IQR \(\approx\) 128). There seems to be little or no linear correlation between completion percentage and duration (Pearson’s \(r\approx\) -0.05).
The first 10 variables of the dataset contain metadata on completion, progress, respondent id, etc.
Data collection: The observations were collected online, using a questionnaire developed for the occasion. [expand on collection method, how was the questionnaire distributed?]
Missing data analysis
To be able to properly take action with respect to gaps in information (missing data), it is paramount to achieve some level of understanding of the underlying process that led to incomplete observations. A useful framework to adopt when discussing missingness was proposed by Rubin (1976), who identified three possible mechanisms responsible for it:
- Missing Completely at Random (MCAR, the probability to be missing is unrelated to observed and unobserved information)
- Missing at Random (MAR, the probability to be missing may depend on observed information)
- Missing Not at Random (MNAR, the probability to be missing may also depend on unobserved information).
Without going into too much detail, Rubin’s framework provides a simple way to assess the validity of many commonly used missing data handling methods, under the three mechanisms above. Broadly speaking, we would like to be dealing with MCAR (hardly ever occurring in practice, but the only situation where listwise and pairwise deletion - as well as mean imputation - do not affect the validity of inference). In practice, however, MAR and MNAR are far more likely situations, both calling for increased care in how missing observations are treated. As no formal statistical testing exists to choose from the three possibilities, inspection of patterns of missingness in the data (also with the aid of visual instruments) paired with expert knowledge is typically the most robust strategy.
The dataset contains 10963 missing observations, which is \(\approx\) 27.69% of the total (435 rows \(\times\) 91 columns \(=\) 39585). Approximately 89.01% of the variables contain at least one missing value, as do 100% of the individual records. If of interest, it is possible to inspect the frequency tables of each of the variables and individuals.
The moderate missingness rate and large number of involved variables and individuals, would suggest that the observations might not be missing in a sparse way. How to get a sense of possible local clusters of missing patterns occurring? A simple way is to produce a visualisation of the whole dataset after recoding it to a missingness indicator matrix (where each values maps to either 0 - not missing - or 1 - missing), possibly sorting the variables from highest to lowest number of missing observations. The next figure contains such visualisation.
Each of the vertical grey ticks indicates one column (variable), whereas each row is a subject. While pretty crowded (we have 91 variables in the dataset), it helps provide an overall look at missingness. What emerges, is that the overall missingness is not low (as observed earlier, ~27.69%), however, it would appear that most of those missing values are specific to a small number of variables and individuals. It is certainly of interest to focus on these, to see if any noticeable pattern arises.
In order to do so, a very useful visual tool is the upset plot. Originally developed to depict intersections between sets (think of it as visualising a contingency table with some additional information on top), it can be used to compactly highlight where the missing observations occur in a dataset, and what combinations of missingness across a set of variables are more frequent. What follows is a missingness upset plot for the 10 variables with the largest amount of missing observations over up to 30 most frequent missingness patterns across them.
The 10 variables considered account for \(\approx\) 27.29% of the total missingness. With almost 100% missingness, Q2_11_TEXT, Q3_12_TEXT, Q8_9_TEXT and Q6_7_TEXT are by large the variables with the highest level of missingness. Inspecting the upset plot reveals that Q11_1_TEXT and Q11_2_TEXT closely follow (77.93%, and 75.63%, respectively), often presenting jointly missing values. Combinations of missing values on all the of the variables above account for a large part of the observed missingness patterns. This is likely due to those variable being referred to the “Other, please specify” option on the relative questions. [double-check]
Reduced dataset: progress at least 81%, no heavily missing field texts (Q2_11_TEXT, etc)
The dataset contains 1061 missing observations, which is \(\approx\) 4% of the total (312 rows \(\times\) 85 columns \(=\) 26520). Approximately 88.24% of the variables contain at least one missing value, as do 29.17% of the individual records. If of interest, it is possible to inspect the frequency tables of each of the variables and individuals.
The relatively low missingness rate and large number of involved variables and individuals, would suggest that the observations are missing in a relatively sparse way.
The overall missingness is low (~4%), however, it would still appear that most of those missing values are specific to a small number of variables and individuals.
What follows is a missingness upset plot for the 10 variables with the largest amount of missing observations over up to 30 most frequent missingness patterns across them.
The upset plot shows how the remaining missingness is almost entirely due a certain number of individuals not having answered to most of the block of Q38 items, more specifically, Q38_4, Q38_6 - Q38_8, Q38_10 - Q38_15. There appears to be no marked differences (not show here) in terms of age (Q1) gender identity (Q2), ethnicity (Q3), [can’t seem to find Q4 in the data?], previous primary responsibility for dependent children (Q4), education (Q6), years of service (Q7), job title (Q8).
Descriptives
The table below allows exploration of the dataset.
In terms of demographics, the tables below contains some summary measures.
Individual variables
Age
## Min. 1st Qu. Median Mean 3rd Qu. Max. NA's
## 22.00 27.00 33.00 35.69 44.00 64.00 55Gender
##
## Female Male Prefer not to say
## 230 177 3
## Prefer to self describe
## 2Ethnicity
##
## Aboriginal and/or Torres Strait Islander
## 1
## Asian
## 2
## Australian
## 382
## European
## 12
## New Zealander
## 7
## North American
## 4
## Prefer not to say
## 1
## Prefer to self describe
## 2Past primary responsibility of children
##
## No Prefer not to say Yes
## 202 4 201Education
##
## Diploma Undergraduate Post-Graduate Other
## 47 274 68 22Years of service as paramedic
##
## I am in my graduate year <5 years 5-10 years
## 5 107 123
## 11-15 years 16-20 years >20 years
## 71 39 66Current job title
##
## ACP CCP
## 330 45
## Non-operational clinician/Manager Other
## 26 11(Some) combinations
Age \(\times\) ethnicity
Gender \(\times\) ethnicity
| Aboriginal and/or Torres Strait Islander | Asian | Australian | European | New Zealander | North American | Prefer not to say | Prefer to self describe | |
|---|---|---|---|---|---|---|---|---|
| Female | 0 | 1 | 216 | 6 | 3 | 2 | 0 | 1 |
| Male | 1 | 1 | 163 | 5 | 4 | 2 | 0 | 1 |
| Prefer not to say | 0 | 0 | 1 | 1 | 0 | 0 | 1 | 0 |
| Prefer to self describe | 0 | 0 | 2 | 0 | 0 | 0 | 0 | 0 |
Gender \(\times\) education
| Diploma | Undergraduate | Post-Graduate | Other | |
|---|---|---|---|---|
| Female | 21 | 169 | 32 | 8 |
| Male | 25 | 102 | 35 | 14 |
| Prefer not to say | 1 | 1 | 1 | 0 |
| Prefer to self describe | 0 | 2 | 0 | 0 |
Gender \(\times\) years of service
| I am in my graduate year | <5 years | 5-10 years | 11-15 years | 16-20 years | >20 years | |
|---|---|---|---|---|---|---|
| Female | 2 | 78 | 74 | 34 | 20 | 22 |
| Male | 3 | 27 | 49 | 35 | 18 | 44 |
| Prefer not to say | 0 | 1 | 0 | 1 | 1 | 0 |
| Prefer to self describe | 0 | 1 | 0 | 1 | 0 | 0 |
education \(\times\) years of service
| I am in my graduate year | <5 years | 5-10 years | 11-15 years | 16-20 years | >20 years | |
|---|---|---|---|---|---|---|
| Diploma | 0 | 2 | 5 | 15 | 12 | 13 |
| Undergraduate | 5 | 98 | 90 | 29 | 16 | 36 |
| Post-Graduate | 0 | 5 | 21 | 17 | 9 | 16 |
| Other | 0 | 2 | 7 | 10 | 2 | 1 |
education \(\times\) current job title
| ACP | CCP | Non-operational clinician/Manager | Other | |
|---|---|---|---|---|
| Diploma | 40 | 1 | 3 | 3 |
| Undergraduate | 244 | 12 | 15 | 3 |
| Post-Graduate | 31 | 27 | 6 | 4 |
| Other | 14 | 5 | 2 | 1 |
Perceived knowledge (Q12_1 - Q12_6)
The dataframe has been reduced to those individuals with no more that 20% of missingness. This results in 291 records being considered from now on.
Overall distribution of perceived knowledge scores.
In relative terms (rounded):
| 1 | 2 | 3 | 4 | 5 | 6 | |
|---|---|---|---|---|---|---|
| Weak | 32 | 7 | 27 | 22 | 39 | 15 |
| Adequate | 65 | 77 | 68 | 69 | 56 | 73 |
| Excellent | 3 | 16 | 4 | 9 | 5 | 12 |
Next, we take a look at how perceived knowledge scores relate to:
Age 
Gender (from lighter to darker: “Female”, “Male”, “Prefer not to say”, “prefer to self describe”) 
Education (from lighter to darker: “Diploma”, “Undergraduate”, “Post-Graduate”, “Other”) 
Years of service (from lighter to darker: “I am in my graduate year”, “<5 years”, “5-10 years”, “11-15 years”, “16-20 years”, “>20 years”) 
Job title (from lighter to darker: “ACP”, “CCP”, “Non-operational clinician/Manager”, “Other”) 
Assessed knowledge (Q13 - Q36)
Overall distribution of assessed knowledge scores.
In relative terms (rounded):
| D1 | D2 | D3 | D4 | D5 | D6 | |
|---|---|---|---|---|---|---|
| Weak | 49 | 65 | 33 | 13 | 83 | 27 |
| Adequate | 41 | 30 | 40 | 33 | 16 | 46 |
| Excellent | 10 | 5 | 27 | 54 | 1 | 27 |
Next, we take a look at how assessed knowledge scores relate to:
Age 
Gender (from lighter to darker: “Female”, “Male”, “Prefer not to say”, “prefer to self describe”) 
Education (from lighter to darker: “Diploma”, “Undergraduate”, “Post-Graduate”, “Other”) 
Years of service (from lighter to darker: “I am in my graduate year”, “<5 years”, “5-10 years”, “11-15 years”, “16-20 years”, “>20 years”)
Job title (from lighter to darker: “ACP”, “CCP”, “Non-operational clinician/Manager”, “Other”)
Perceived vs assessed knowledge
In the following, r_s denotes Spearman’s correlation
Q37_1 versus assessed knowledge
D1: Antepartum complications
##
## Weak Adequate Excellent
## Strongly disagree 17 24 4
## Somewhat disagree 43 27 9
## Neither agree nor disagree 15 13 3
## Somewhat agree 53 46 8
## Strongly agree 15 9 4
D2: Physiological birth
##
## Weak Adequate Excellent
## Strongly disagree 30 12 3
## Somewhat disagree 55 19 5
## Neither agree nor disagree 21 10 0
## Somewhat agree 68 32 7
## Strongly agree 14 13 1
D3: Intrapartum complications
##
## Weak Adequate Excellent
## Strongly disagree 13 18 14
## Somewhat disagree 30 33 16
## Neither agree nor disagree 10 13 8
## Somewhat agree 36 40 31
## Strongly agree 7 12 9
D4: Postpartum complications
##
## Weak Adequate Excellent
## Strongly disagree 6 17 22
## Somewhat disagree 14 28 37
## Neither agree nor disagree 6 11 14
## Somewhat agree 8 32 67
## Strongly agree 3 8 17
D5: Trauma in pregnancy
##
## Weak Adequate Excellent
## Strongly disagree 36 9 0
## Somewhat disagree 69 10 0
## Neither agree nor disagree 26 5 0
## Somewhat agree 89 17 1
## Strongly agree 20 7 1
Q37_4 versus assessed knowledge
D1: Antepartum complications
##
## Weak Adequate Excellent
## Strongly disagree 12 9 3
## Somewhat disagree 28 37 5
## Neither agree nor disagree 46 17 8
## Somewhat agree 50 50 10
## Strongly agree 7 6 2
D2: Physiological birth
##
## Weak Adequate Excellent
## Strongly disagree 17 5 2
## Somewhat disagree 47 20 3
## Neither agree nor disagree 43 23 5
## Somewhat agree 71 33 6
## Strongly agree 11 4 0
D3: Intrapartum complications
##
## Weak Adequate Excellent
## Strongly disagree 10 7 7
## Somewhat disagree 22 32 16
## Neither agree nor disagree 28 25 18
## Somewhat agree 32 45 33
## Strongly agree 4 6 5
D4: Postpartum complications
##
## Weak Adequate Excellent
## Strongly disagree 2 7 15
## Somewhat disagree 10 20 40
## Neither agree nor disagree 14 31 26
## Somewhat agree 11 34 65
## Strongly agree 0 4 11
D5: Trauma in pregnancy
##
## Weak Adequate Excellent
## Strongly disagree 20 3 1
## Somewhat disagree 60 10 0
## Neither agree nor disagree 60 11 0
## Somewhat agree 88 21 1
## Strongly agree 12 3 0
Q37_5 versus assessed knowledge
D1: Antepartum complications
##
## Weak Adequate Excellent
## Strongly disagree 13 13 2
## Somewhat disagree 24 28 6
## Neither agree nor disagree 42 17 6
## Somewhat agree 58 53 13
## Strongly agree 7 8 1
D2: Physiological birth
##
## Weak Adequate Excellent
## Strongly disagree 20 5 3
## Somewhat disagree 41 15 2
## Neither agree nor disagree 34 28 3
## Somewhat agree 82 34 8
## Strongly agree 12 4 0
D3: Intrapartum complications
##
## Weak Adequate Excellent
## Strongly disagree 8 12 8
## Somewhat disagree 23 24 11
## Neither agree nor disagree 25 27 13
## Somewhat agree 37 46 41
## Strongly agree 3 7 6
D4: Postpartum complications
##
## Weak Adequate Excellent
## Strongly disagree 2 9 17
## Somewhat disagree 8 18 32
## Neither agree nor disagree 14 26 25
## Somewhat agree 13 42 69
## Strongly agree 0 2 14
D5: Trauma in pregnancy
##
## Weak Adequate Excellent
## Strongly disagree 23 4 1
## Somewhat disagree 49 9 0
## Neither agree nor disagree 60 5 0
## Somewhat agree 97 26 1
## Strongly agree 12 4 0
Q37_12 versus assessed knowledge
D1: Antepartum complications
##
## Weak Adequate Excellent
## Strongly disagree 9 6 0
## Somewhat disagree 42 20 7
## Neither agree nor disagree 44 32 6
## Somewhat agree 37 47 10
## Strongly agree 12 14 5
D2: Physiological birth
##
## Weak Adequate Excellent
## Strongly disagree 12 3 0
## Somewhat disagree 45 21 3
## Neither agree nor disagree 57 21 4
## Somewhat agree 59 31 4
## Strongly agree 16 10 5
D3: Intrapartum complications
##
## Weak Adequate Excellent
## Strongly disagree 4 10 1
## Somewhat disagree 25 30 14
## Neither agree nor disagree 34 28 20
## Somewhat agree 23 36 35
## Strongly agree 10 12 9
D4: Postpartum complications
##
## Weak Adequate Excellent
## Strongly disagree 4 5 6
## Somewhat disagree 12 25 32
## Neither agree nor disagree 13 26 43
## Somewhat agree 6 34 54
## Strongly agree 2 7 22
D5: Trauma in pregnancy
##
## Weak Adequate Excellent
## Strongly disagree 13 2 0
## Somewhat disagree 65 4 0
## Neither agree nor disagree 69 13 0
## Somewhat agree 73 19 2
## Strongly agree 21 10 0
Q37_13 versus assessed knowledge
D1: Antepartum complications
##
## Weak Adequate Excellent
## Strongly disagree 8 2 1
## Somewhat disagree 43 26 6
## Neither agree nor disagree 43 27 6
## Somewhat agree 38 54 11
## Strongly agree 12 10 4
D2: Physiological birth
##
## Weak Adequate Excellent
## Strongly disagree 9 2 0
## Somewhat disagree 51 21 3
## Neither agree nor disagree 53 20 3
## Somewhat agree 61 36 6
## Strongly agree 15 7 4
D3: Intrapartum complications
##
## Weak Adequate Excellent
## Strongly disagree 2 7 2
## Somewhat disagree 27 31 17
## Neither agree nor disagree 30 30 16
## Somewhat agree 28 38 37
## Strongly agree 9 10 7
D4: Postpartum complications
##
## Weak Adequate Excellent
## Strongly disagree 4 4 3
## Somewhat disagree 15 26 34
## Neither agree nor disagree 11 26 39
## Somewhat agree 5 35 63
## Strongly agree 2 6 18
D5: Trauma in pregnancy
##
## Weak Adequate Excellent
## Strongly disagree 10 1 0
## Somewhat disagree 69 6 0
## Neither agree nor disagree 62 14 0
## Somewhat agree 83 18 2
## Strongly agree 17 9 0
Q38_3 versus assessed knowledge
D1: Antepartum complications
##
## Weak Adequate Excellent
## Strongly disagree 16 5 2
## Somewhat disagree 32 31 6
## Neither agree nor disagree 34 24 6
## Somewhat agree 54 48 11
## Strongly agree 2 7 1
D2: Physiological birth
##
## Weak Adequate Excellent
## Strongly disagree 17 4 2
## Somewhat disagree 45 20 4
## Neither agree nor disagree 39 21 4
## Somewhat agree 76 33 4
## Strongly agree 6 3 1
D3: Intrapartum complications
##
## Weak Adequate Excellent
## Strongly disagree 7 10 6
## Somewhat disagree 20 30 19
## Neither agree nor disagree 25 20 19
## Somewhat agree 37 46 30
## Strongly agree 1 4 5
D4: Postpartum complications
##
## Weak Adequate Excellent
## Strongly disagree 1 9 13
## Somewhat disagree 12 23 34
## Neither agree nor disagree 12 21 31
## Somewhat agree 11 38 64
## Strongly agree 0 3 7
D5: Trauma in pregnancy
##
## Weak Adequate Excellent
## Strongly disagree 18 4 1
## Somewhat disagree 56 12 1
## Neither agree nor disagree 59 5 0
## Somewhat agree 91 22 0
## Strongly agree 8 2 0
Q38_4 versus assessed knowledge
D1: Antepartum complications
##
## Weak Adequate Excellent
## Strongly disagree 20 7 2
## Somewhat disagree 38 26 3
## Neither agree nor disagree 27 25 7
## Somewhat agree 50 50 12
## Strongly agree 3 7 2
D2: Physiological birth
##
## Weak Adequate Excellent
## Strongly disagree 21 7 1
## Somewhat disagree 44 19 4
## Neither agree nor disagree 37 19 3
## Somewhat agree 77 30 5
## Strongly agree 4 6 2
D3: Intrapartum complications
##
## Weak Adequate Excellent
## Strongly disagree 9 13 7
## Somewhat disagree 16 32 19
## Neither agree nor disagree 22 19 18
## Somewhat agree 39 42 31
## Strongly agree 4 4 4
D4: Postpartum complications
##
## Weak Adequate Excellent
## Strongly disagree 4 11 14
## Somewhat disagree 12 24 31
## Neither agree nor disagree 10 19 30
## Somewhat agree 10 33 69
## Strongly agree 0 6 6
D5: Trauma in pregnancy
##
## Weak Adequate Excellent
## Strongly disagree 24 4 1
## Somewhat disagree 56 10 1
## Neither agree nor disagree 55 4 0
## Somewhat agree 88 24 0
## Strongly agree 9 3 0
Q38_2 versus Q38_9
##
## Strongly disagree Somewhat disagree
## Strongly disagree 14 3
## Somewhat disagree 22 30
## Neither agree nor disagree 35 21
## Somewhat agree 72 38
## Strongly agree 7 3
##
## Neither agree nor disagree Somewhat agree
## Strongly disagree 0 0
## Somewhat disagree 6 2
## Neither agree nor disagree 4 3
## Somewhat agree 9 4
## Strongly agree 0 0
##
## Strongly agree
## Strongly disagree 1
## Somewhat disagree 2
## Neither agree nor disagree 2
## Somewhat agree 1
## Strongly agree 1
Q38_8 versus Q37_4
##
## Strongly disagree Somewhat disagree
## Strongly disagree 1 3
## Somewhat disagree 3 7
## Neither agree nor disagree 2 13
## Somewhat agree 12 27
## Strongly agree 6 14
##
## Neither agree nor disagree Somewhat agree
## Strongly disagree 3 2
## Somewhat disagree 9 13
## Neither agree nor disagree 16 12
## Somewhat agree 30 48
## Strongly agree 11 32
##
## Strongly agree
## Strongly disagree 1
## Somewhat disagree 2
## Neither agree nor disagree 2
## Somewhat agree 3
## Strongly agree 6
Q38_8 versus Q37_1
##
## Strongly disagree Somewhat disagree
## Strongly disagree 2 4
## Somewhat disagree 3 11
## Neither agree nor disagree 8 13
## Somewhat agree 16 33
## Strongly agree 14 14
##
## Neither agree nor disagree Somewhat agree
## Strongly disagree 1 2
## Somewhat disagree 3 13
## Neither agree nor disagree 2 16
## Somewhat agree 16 46
## Strongly agree 6 28
##
## Strongly agree
## Strongly disagree 1
## Somewhat disagree 4
## Neither agree nor disagree 6
## Somewhat agree 9
## Strongly agree 7