Exploratory data analysis
DFU DCC: Dr. Spino, Dr. Song, Soumik.
2022-06-30
Table 1 for demographic covariates
Click here for more details.
| Indiana (N=6) |
Stanford (N=6) |
UCSF (N=18) |
UMich (N=30) |
Overall (N=60) |
|
|---|---|---|---|---|---|
| Sex | |||||
| Female | 3 (50.0%) | 1 (16.7%) | 4 (22.2%) | 3 (10.0%) | 11 (18.3%) |
| Male | 3 (50.0%) | 5 (83.3%) | 14 (77.8%) | 27 (90.0%) | 49 (81.7%) |
| Age (years) | |||||
| Mean (SD) | 54.0 (11.4) | 70.0 (9.53) | 55.3 (8.83) | 56.0 (12.6) | 57.0 (11.8) |
| Median [Min, Max] | 55.0 [34.0, 67.0] | 69.5 [58.0, 81.0] | 58.0 [40.0, 70.0] | 60.0 [29.0, 76.0] | 59.0 [29.0, 81.0] |
| Highest level of education | |||||
| Bachelor's degree (e.g., BA, AB, BS, BBA) | 1 (16.7%) | 2 (33.3%) | 2 (11.1%) | 2 (6.7%) | 7 (11.7%) |
| High school graduate | 2 (33.3%) | 0 (0%) | 3 (16.7%) | 3 (10.0%) | 8 (13.3%) |
| Master's degree (e.g., MA, MS, MEng, MEd, MBA) | 1 (16.7%) | 0 (0%) | 1 (5.6%) | 4 (13.3%) | 6 (10.0%) |
| Unknown | 2 (33.3%) | 3 (50.0%) | 0 (0%) | 0 (0%) | 5 (8.3%) |
| Some college, no degree | 0 (0%) | 1 (16.7%) | 8 (44.4%) | 15 (50.0%) | 24 (40.0%) |
| Associate degree: academic program | 0 (0%) | 0 (0%) | 1 (5.6%) | 0 (0%) | 1 (1.7%) |
| Doctoral degree (e.g., PhD, EdD) | 0 (0%) | 0 (0%) | 1 (5.6%) | 2 (6.7%) | 3 (5.0%) |
| No diploma | 0 (0%) | 0 (0%) | 2 (11.1%) | 0 (0%) | 2 (3.3%) |
| Associate degree: occupational or technical or vocational program | 0 (0%) | 0 (0%) | 0 (0%) | 3 (10.0%) | 3 (5.0%) |
| GED or equivalent | 0 (0%) | 0 (0%) | 0 (0%) | 1 (3.3%) | 1 (1.7%) |
| Employment status | |||||
| Disabled, permanently or temporarily | 3 (50.0%) | 0 (0%) | 7 (38.9%) | 10 (33.3%) | 20 (33.3%) |
| Looking for work, unemployed | 1 (16.7%) | 0 (0%) | 1 (5.6%) | 1 (3.3%) | 3 (5.0%) |
| Retired | 1 (16.7%) | 4 (66.7%) | 5 (27.8%) | 10 (33.3%) | 20 (33.3%) |
| Working now | 1 (16.7%) | 2 (33.3%) | 4 (22.2%) | 7 (23.3%) | 14 (23.3%) |
| Laid off | 0 (0%) | 0 (0%) | 1 (5.6%) | 0 (0%) | 1 (1.7%) |
| Other | 0 (0%) | 0 (0%) | 0 (0%) | 1 (3.3%) | 1 (1.7%) |
| Sick leave or maternity leave | 0 (0%) | 0 (0%) | 0 (0%) | 1 (3.3%) | 1 (1.7%) |
| Marital status | |||||
| Divorced | 1 (16.7%) | 0 (0%) | 3 (16.7%) | 3 (10.0%) | 7 (11.7%) |
| Married | 1 (16.7%) | 3 (50.0%) | 3 (16.7%) | 15 (50.0%) | 22 (36.7%) |
| Never married | 2 (33.3%) | 2 (33.3%) | 9 (50.0%) | 7 (23.3%) | 20 (33.3%) |
| Widowed | 2 (33.3%) | 1 (16.7%) | 0 (0%) | 4 (13.3%) | 7 (11.7%) |
| Domestic partner | 0 (0%) | 0 (0%) | 2 (11.1%) | 1 (3.3%) | 3 (5.0%) |
| Separated | 0 (0%) | 0 (0%) | 1 (5.6%) | 0 (0%) | 1 (1.7%) |
Percentage of missingness
Click here for more details.
For patient \(j\) in center \(i\), we calculate the
Percentage (%) of missingness denoted by \(P_{ij}\) as follows: \[P_{ij} = 100 \times \left(\frac{\text{# of SOC
readings} - \text{# of eKARE readings}}{\text{# of SOC readings}}
\right).\]
Post-matching summary
Click here for more details.
Patient-level percentage error for measurements
Click here for more details.
In this section we assume SOC is the gold-standard and
compare accuracy of eKARE measurements relative to
SOC measurements.
For patient \(j\) in center \(i\) and visit \(k\), we calculate the
absolute percentage error for a measurement type (say,
width) denoted by \(APE^{w}_{ijk}\) as follows: \[APE^{w}_{ijk} = 100 \times \left(\frac{\left|
\text{SOC}^w_{ijk} - \text{eKARE}^w_{ijk} \right|}{ \text{SOC}^w_{ijk}}
\right).\] We summarise this percentage error at a patient-level
by computing \(MAPE_{ij}^w =
\text{mean}(APE_{ijk}^w)\) and examine boxplots for each of the
three measurement categories: length, width,
and area.
Table 2 for summary of measurement error
Click here for more details.
| Indiana (N=6) |
Stanford (N=6) |
UCSF (N=18) |
UMich (N=30) |
Overall (N=60) |
|
|---|---|---|---|---|---|
| Age (years) | |||||
| Mean (SD) | 54.0 (11.4) | 70.0 (9.53) | 55.3 (8.83) | 56.0 (12.6) | 57.0 (11.8) |
| Median [Min, Max] | 55.0 [34.0, 67.0] | 69.5 [58.0, 81.0] | 58.0 [40.0, 70.0] | 60.0 [29.0, 76.0] | 59.0 [29.0, 81.0] |
| Gender | |||||
| Female | 3 (50.0%) | 1 (16.7%) | 4 (22.2%) | 3 (10.0%) | 11 (18.3%) |
| Male | 3 (50.0%) | 5 (83.3%) | 14 (77.8%) | 27 (90.0%) | 49 (81.7%) |
| Length MAPE (%) | |||||
| Mean (SD) | 77.0 (58.0) | 40.3 (44.1) | 21.6 (15.7) | 36.6 (43.1) | 36.6 (40.9) |
| Median [Min, Max] | 83.0 [2.38, 162] | 23.1 [14.3, 130] | 23.0 [0, 58.8] | 23.6 [0, 169] | 23.9 [0, 169] |
| Width MAPE (%) | |||||
| Mean (SD) | 129 (117) | 51.7 (68.5) | 31.5 (33.2) | 47.8 (44.4) | 51.4 (59.9) |
| Median [Min, Max] | 74.0 [46.3, 340] | 23.4 [5.00, 187] | 28.0 [0, 150] | 33.7 [0, 194] | 31.7 [0, 340] |
| Area MAPE (%) | |||||
| Mean (SD) | 256 (276) | 50.8 (40.2) | 65.7 (59.7) | 65.0 (87.0) | 82.9 (121) |
| Median [Min, Max] | 139 [19.5, 618] | 35.5 [14.3, 111] | 43.8 [16.0, 202] | 35.4 [9.32, 347] | 37.4 [9.32, 618] |
Questions
- What models should I consider?
- Model to explain missingness? we have patient-level data on missingness % and demographic information. Are significant predictors of missingness present?
- Model to study error in match? for each of the three measurement types, have error values for measurements - mixed models?
- What other graphics should I look at?