Out of the 5476 registered patients, 3183 (58.13%) were actively following their clinic appointments, while 2293 (41.87%) were inactive on 30/12/2014, 366 of the inactive group were reported dead (Figure 1). 36.34% were children under five years, 45.38% were 5-17 years old while children under one year were 402 (7.34%). 2797 (51.08%) were males, while females were 2679 (48.92%) (Table 1). Most of the enrolled patients (72.92%) were born in the Coastal regions such as Dar es Salaam and Tanga.
Table 1: Demographics characteristics of patients enrolled in the Muhimbili Sickle Cohort
|
Total
|
All SCD
|
5476
|
Age groups
|
|
0-4
|
2294 (41.89%)
|
5-17
|
2485 (45.38%)
|
18 and over
|
680 (12.42%)
|
Gender
|
|
Female
|
2679 (48.92%)
|
Male
|
2797 (51.08%)
|
Place of Birth
|
|
Coastal Regions
|
3993(72.92%)
|
Others
|
1476 (26.95%)
|
Missing
|
7 (0.13%)
|
The trends in enrollment and LTFU patterns from 2004 to 2014 are shown in Figure 2. The highest numbers of registrations were achieved at the beginning of the study in 2004, while the LTFU event increases with time. The number of LTFU events (Figure 2) in any particular year is not proportional to the number of patients registered in that year since a patient may be LTFU in a different year to the one in which they were registered. Clinic attendance up to 2014 included, 2015-2016 was excluded as additional sickle cell clinics were opened at the government referral hospitals to which SCD patients were referred. Therefore, they cannot be categorized as LTFU as they might be accessing healthcare services elsewhere. But patients register in 2015 were 645 while in 2016 were 170. Table 2 compares the clinical and laboratory data from active patients those who are not LTFU with those who were LTFU (inactive group). This table shows the median values, interquartile range and p-value from the T-test statistics. Based on the p-values obtained there are statistically significant differences between the two groups in age, mean cell volume (MCV) and mean cell hemoglobin concentration (MCHC). The median age of active patients (NOT LTFU) was 13 years compared to 11 years for inactive patients (LTFU). The level of MCV was slightly higher for those patients who were actively following their clinic appointment compared to those who were not.
Table 2: Comparison of laboratory and clinical parameters from active (NOT LTFU) and inactive (LTFU) patients.
|
Active patients (NOT LTFU)
|
Inactive patients(LTFU exclude those reported dead)
|
|
|
Median
|
IQR
|
Median
|
IQR
|
Test Statistics
|
Age
|
13.00
|
07.00-19.00
|
11.00
|
07.00-15.00
|
t = 5.696 (p<0.001)
|
White blood cell (WBC)
(10^3/uL)
|
13.11
|
10.69-16.60
|
13.65
|
10.90-17.19
|
t=-0.9953
(0.3197)
|
Neutrophils (%)
|
46.50
|
38.80-54.52
|
45.20
|
38.10-53.25
|
t=1.9524
(p=0.0509)
|
Hemoglobin (g/dL)
|
07.40
|
06.60-08.30
|
07.40
|
6.70-08.20
|
t=0.0620
(p=0.9505)
|
Red Blood Cells (RBC)
(10^6/uL)
|
02.85
|
02.42-03.33
|
02.83
|
2.44-3.28
|
t=0.1212
(p=0.9036)
|
Mean Cell Volume (MCV)
(fL)
|
78.58
|
72.00-85.20
|
77.40
|
71.50-83.80
|
t=2.3742
(p=0.0177)
|
Mean Cell Hemoglobin Concentration (MCHC)(g/dL)
|
33.33
|
31.60-34.90
|
33.80
|
32.27-35.10
|
t=-4.0359
(p<0.001)
|
Red Cell Distribution Width (RDW)
(%)
|
22.00
|
19.70-24.40
|
22.30
|
19.80-24.70
|
t=-0.2216
(p=0.8247)
|
Platelets (10^3/uL)
|
417.00
|
304.00-522.00
|
409.50
|
312 -522
|
t=-0.0655
(p=0.9478)
|
Haematocrit (%)
|
22.22
|
19.70-22.20
|
22.80
|
19.60-24.50
|
t = 1.5341
(p=0.1251)
|
Survival analysis is used to model survival time or the time until the event of interest. For our case, our event was LTFU. Figure 3A shows the survival probability of SCD patients over time and cumulative LTFU events over time (Figure 3B) with a 95% confidence interval using the Kaplan-Meir method. At the beginning of the study, the survival probability was one which is expected since no LTFU events have occurred. The curve is horizontal up to around 300 days (> 9 months). The median survival time was 2848 days (7.8 years), with a 95% confidence interval of [2716-3004] days. By 2013, ten years from the beginning of the MSC, the survival probability is around 0.2113 (21.13%), which means a retention rate of 78.87%.
In addition to the survival trends, we compared the survival probabilities for different groups using the LogRank test and Cox hazard analysis (with a significant level of 5%). A univariate and multivariate Cox hazard analysis were implemented to select significant attributes (Table 4). Figure 4 shows survival curves for these four groups to the time LTFU events occurred. Figure 4A shows a comparison between three age groups categories demonstrating that patients who are 18 and over are least likely to be LTFU, followed by those who are 5-17 years old, with children under five the highest proportion of LTFU (p<0.001). We also stratified patients based on fever, painful episodes and well today (the well variable (Figure 4D) was used to record whether the patient was feeling well or not on that day). Patients who experience painful episodes (Figure 4C) were the least likely to be LTFU compared to those with no painful episodes (p=0.00052). Patients with fever (Figure 4B) were more likely to be LTFU than those with no fever (p=0.035). Table 3 shows the hazard ratios from the Cox proportional hazard model, with the age group patients who are 18 and over used as a reference group. The hazard of patients aged 5-17 years is 2.61 (95% CI: 2.2324 - 3.0705, p< 0.001) times higher than those 18 and over, while for children under five, it is 14.29 times higher (95% CI: 11.0071-18.5768, p <0.001) than those 18 and over.
In the Cox model, the painful and fever variables were removed due to lack of independence with the age group and well today attributes. Using the chi-square test, age group and painful episodes, attributes were found to depend on one another with a p-value of 8.889e-07 and were fever and well today (p= 5.96e-07) interdependent. Therefore, survival curves are shown, but only the hazard ratio result for well today. The hazard ratio of those who are well was 13.31 (95% CI: 1.0003-1.2837, p=0.0495) higher than those who are not well-meaning those who felt well were more likely to be LTFU than those who were feeling unwell. The hazard ratio for patients with white blood cell (WBC) counts above average (14.34) was 1.58 (95% CI:1.0042 - 1.0276, p=0.0074) higher than those below average, indicating those with an above-average WBC count are more likely to be LTFU than those with lower WBC. Patients with hematocrit levels above average value (22.69) have a hazard ratio of 2.38 (95% CI: 1.0076 - 1.0404, p=0.0039) higher than those with a value below average, indicating those with above-average hematocrit values are more likely to be LTFU.
Table 3: Final results of the Cox proportional hazard model with significant attributes
Attribute
|
Estimate
|
Hazard ratio
|
P-value
|
95 % CI
|
White blood cells (10^3/uL)
|
0.0157
|
1.0158
|
0.0074
|
1.0042-1.0276
|
Hematocrit (%)
|
0.0235
|
1.0238
|
0.0039
|
1.0076-1.0404
|
Well today (Yes )
|
0.1249
|
1.1331
|
0.0495
|
1.0003-1.2837
|
Age group (5-17)
|
0.9624
|
2.6181
|
< 0.001
|
2.2324-3.0705
|
Age group (<5)
|
2.6602
|
14.2995
|
<0.001
|
11.0071-18.5768
|
mch : time group (0-2050 days)
|
-0.1286
|
0.8792
|
<0.001
|
0.8403-0.9200
|
mch : time group (>2050)
|
0.0073
|
1.0073
|
0.8265
|
0.9432-1.0759
|
mcv : time group (0-2050 days)
|
0.0419
|
1.0428
|
<0.001
|
1.026-1.0598
|
mcv : time group (>2050)
|
-0.0005
|
0.9994
|
0.9629
|
0.9745-1.0249
|
During the model, fitting mean corpuscular hemoglobin (MCH) and mean cell volume (MCV) were found to be highly significant but violating proportionality assumptions. To rectify this, the stratification of their variable was performed by splitting the follow-up period. Table 5 in the appendix shows the results of proportionality assumptions after stratification, so all attributes are satisfying this assumption. We considered two follow-up intervals, i) 0-2050 days and ii) 2050 to the end of the study. The effect of MCV and MCH on LTFU is limited to the first period (0-2050 days). Patients with MCV value above average (77.73) have a hazard ratio of 4.28 (95% CI: 1.0260 - 1.0598, p<0.001) higher than those below average and hence patients with MCV above average are more likely to be LTFU than those below the average value. Patients with MCH values higher than average value have a hazard ratio of 12(95% CI: 0.8403 - 0.9200, p<0.001) less than those below the average value, meaning those with lower MCH values are less likely to be LTFU (Table 3).