Trend and Correlates of Stillbirth, Maternal and Neonatal Deaths in the Eastern Region of Ghana

Author

Kwabena Asare

Published

November 4, 2024

Introduction

PURPOSE: To conduct statistical analysis to investigate associations between;

  • Exposures: Health Facility service delivery variables (antenatal care, post natal care, IPT receipt, ITN distribution and IFA).

  • Outcomes: Maternal mortality, Stillbirth, Low birth weight, Neonatal mortality

    • Note: It is conceptually better to consider low birth weight as an outcome instead of as an exposure, particularly for the neonatal mortality. LBW is a mediator for the relationship between any exposure and neonatal death and hence should not be adjusted for in such models. Also from a health service angle, normal birth weight is a goal/objective/target because we know for sure that low birth weight is the main cause of neonatal death.

APPROACH: Explore each exposure with each outcome building models with a set of confounders carefully selected based on the type of exposure-outcome in question.

DATA RECEIVED (STRUCTURE)

All data (in counts) are aggregated by health facility type (listed below): Thus the unit of analysis is Health Facility type and not health facility.

  • CHPS compounds.

  • Clinics.

  • Health centers.

  • District hospital and Other hospital.

  • Regional hospital (Data came separately). Data for the rest of the health facilities are together.

This is the biggest issue with this whole project.

  • 1) Thus in terms of samples size we technically have 5 observations and then if you add the yearly data you have times 12 years each making 60. A sample size of 60 is too small for a meaningful analysis. This affects precision and the reason most of the previous results were not statistically significant.
  • 2) This aggregation of the data makes it difficult to do proper to check for data quality and any bias particularly due to missing data, which is very important in this context. In the previous feedback, I mentioned that most of the CHPS compounds do not perform deliveries in order to record still birth, low birth weight. A health facility level data will be best to check for data quality. The aggregation hides a lot of granularity and bias.
  • 3) All the variables in this analysis occur at the health facility level, thus the unit of analysis should be health facility. Each health facility should be a data point in the data set. That is the most suitable data set for this analysis. But currently that’s not what we have.

Data are longitudinal from 2012 t0 2023.

File names and descriptions of data sets received are below:

  • ANC 1: ANC Registrants, ANC attendance, Mother’s age at registration, Mothers making 4th ANC visit, ITN Distributed, IPT 1 to 4.
  • ANC 2: ITN Distributed, Hemoglobin checked at ANC registration, Hemoglobin level at >=36 weeks of pregnancy, Postpartum IFA Supplementation given, women given IFA for 3 times during pregnancy , women given IFA for 3 times during pregnancy, Pregnant women tested for HIV, Pregnant women tested HIV positive, Number of new positives put on arv, Mothers on arv.
  • Deliveries: Live births, Spontaneous Vaginal Delivery, Vacuum deliveries, Cesarean section deliveries, Parity 1-5, 5+, Newborn with low birth weighs (< 2500 g), Babies with birth weight ≥ 2.5 kg, Delivery by mother in age group.
  • PNC: 1st PNC from day 8 and above, 1st PNC on day 1 or 2 (Babies), 1st PNC on day 3-7, 2nd PNC (day 6-7) (Babies), 2nd PNC (day 6-7) (Mothers), 3rd PNC (at 6weeks) (Mothers), Attendees for 2nd PNC, Total PNC registrants, Breastfeeding initiated within 30 minutes, Mothers initiating breastfeeding within 1Hour of delivery, Exclusive breastfeeding at discharge, Baby weight within 6-10 day < 2.5 kg.
  • Deaths: Total maternal deaths, Abortions, Deaths from post abortions complications, Total Still Birth, Total Neonatal Deaths,Neonatal deaths (<1 month), Newborn with low birth weight (less than 2.5 kg).

Data management.

Tidy data sets by converting as below.

  • Each health facility type and year (2012-2023) as an observation/row.
  • Each variable as a column.
  • Key is to check for missing rates and to determine any possible data management issues and appropriateness of data set for the analysis.

Descriptive analysis and results.

The descriptive results

  1. Generate the descriptive stats and compare with previous analysis if there has been a big change in sample size.
  2. Check the rate of missing variables to determine appropriateness for the analysis

Outcomes from 2012 to 2023

  1. Total Live Births:

  2. Total Stillbirths:

  3. Low birth weight: It is difficult to identify this variable from the data set, so i have provided all possible variables with the words low or weight.

    • newborn_with_low_birth_weight_less_than_2_5_kg_deaths
    • newborn_with_low_birth_weighs_2500_g_cases
    • babies_with_birth_weight_2_5_kg
    • birth_weight_2_5_kg_multipara
    • birth_weight_2_5_kg_primipara
    • low_birth_weight_1_5kg_to_2_499kg
    • newborn_with_low_birth_weighs_2500_g_cases_per_1000_LB
    • The newborn_with_low_birth_weighs_2500_g_cases looks the closest but there is about 3000 difference per year.
    • So which variable is for low_birth_weight or how did you define it.

  4. Total Maternal Deaths:

  5. Total Neonatal Deaths: Why is it missing in 2012 and 2013? The main issue is that NM increases very surprisingly overtime. There is also a sharp increase from 2016 to 2017. Really need to be sure this is not a data management or database/DHIMS error. The answer to this is to get health facility level data.

  6. I have provided trends in the outcomes by facility type to see any consistency with the cumulative trends. You see that this rising total_neonatal_death is driven by the hospitals which can indicate a data error. Might be one hospital with a data issue issue driving this. One could argue that the deaths are likely to be recorded at the hospital since mothers with complications are likely to be sent there, but why the jump from 2016. Also if this is the case, the numbers from the other health facilities should have reduced over time.

  7. So need to identify what is happening since the trend from the other health facility types is the perfect reflection of expected neonatal mortality.

Table 1. Descriptive outcomes from 2012 to 2023
Characteristic 2012, N = 11 2013, N = 11 2014, N = 11 2015, N = 11 2016, N = 11 2017, N = 11 2018, N = 11 2019, N = 11 2020, N = 11 2021, N = 11 2022, N = 11 2023, N = 11 Overall, N = 121
live_births 61,774 58,764 63,467 57,719 58,092 57,442 62,010 62,911 65,419 68,858 68,398 66,163 751,017
total_still_birth 1,152 1,167 1,087 1,054 968 855 885 892 906 960 862 884 11,672
total_still_birth_per_1000_LB 98 98 89 90 91 75 77 75 70 78 74 70 985
babies_with_birth_weight_2_5_kg 53,505 51,731 52,406 51,546 52,894 54,331 57,010 57,238 60,336 64,457 63,309 61,302 680,065
low_birth_weight_1_5kg_to_2_499kg missing missing missing missing missing missing missing missing missing missing missing missing missing
newborn_with_low_birth_weighs_2500_g_cases 406 7 830 902 1,007 1,047 1,014 1,062 1,041 1,152 1,666 942 11,076
newborn_with_low_birth_weighs_2500_g_cases_per_1000_LB 37 1 50 63 69 125 85 72 53 58 84 52 748
total_maternal_deaths 116 124 117 106 97 106 80 91 97 80 85 68 1,167
total_maternal_deaths_per_1000_LB 12.67 15.00 12.72 12.09 12.35 11.99 9.47 10.71 13.10 9.01 11.12 7.97 138.21
total_neonatal_death missing missing 232 221 237 493 429 422 553 468 546 427 4,028
total_neonatal_death_per_1000_LB missing missing 33 33 34 55 47 51 65 57 55 51 481
1 Sum

Risk factors from 2012 to 2023

The number of the below predictors/risk factors are shown in Table 2.

  1. Mothers age at ANC registration.
  2. mothers_making_4th_anc_visit: This looks fine.
  3. IFA supplementation: this is missing from 2012 to 2016. Did the policy begin in 2017? If yes, we can do an interrupted time series analysis to estimate change in outcomes before and after 2017.
  4. itn_distributed: would love to include this but the 2012 and 2015 data looks suspicious.
  5. IPT: Looks fine.
  6. child_on_arv: cannot find variable
  7. mothers_on_arv: missing in 2021 nad 2022 with suspiciuos number in 2023.
  8. total_pnc_registrants: PNC registrants and PNC on day 1/2 looks fine. The other PNC variables do not look good and will be discarded.
  9. exclusive_breastfeeding_at_discharge: looks fine.
  10. Spontaneous abortions: cant find variable
Table 2. Descriptive predictors from 2012 to 2023
Characteristic 2012, N = 11 2013, N = 11 2014, N = 11 2015, N = 11 2016, N = 11 2017, N = 11 2018, N = 11 2019, N = 11 2020, N = 11 2021, N = 11 2022, N = 11 2023, N = 11 Overall, N = 121
anc_registrants 85,282 83,979 83,519 80,653 80,111 78,301 78,917 78,210 82,431 80,782 79,129 73,612 964,926
antenatal_mother_at_registration_10_14 267 270 344 270 243 277 277 317 325 359 321 311 3,581
antenatal_mother_at_registration_15_19 11,935 12,173 11,404 11,293 11,036 10,537 10,032 9,944 9,861 9,875 9,472 8,453 126,015
antenatal_mother_at_registration_20_24 23,378 22,826 22,039 21,337 20,200 19,046 19,242 18,361 19,389 18,882 18,823 17,411 240,934
antenatal_mother_at_registration_25_29 23,302 23,004 21,644 21,690 21,347 20,809 21,114 21,137 21,657 21,197 20,342 18,536 255,779
antenatal_mother_at_registration_30_34 16,136 15,899 15,879 15,939 16,479 16,459 16,538 16,633 18,007 17,489 17,383 16,747 199,588
antenatal_mother_at_registration_35_39 10,400 9,860 9,731 10,266 10,393 10,595 11,657 11,858 13,147 12,897 12,785 12,219 135,808
antenatal_mother_at_registration_40 missing missing missing missing missing missing missing missing missing missing missing missing missing
mothers_making_4th_anc_visit 65,007 57,715 56,703 54,304 57,645 54,144 53,245 54,081 56,554 62,401 63,335 62,693 697,827
itn_distributed 75 2,635 8,826 339 38,339 71,380 76,038 75,082 79,623 76,957 72,682 71,808 573,784
number_of_women_given_ifa_for_3_times_during_pregnancy missing missing missing missing missing 40,988 84,683 68,634 64,700 73,344 73,031 70,710 476,090
number_of_women_given_ifa_for_6_times_during_pregnancy missing missing missing missing missing 22,622 57,608 36,393 42,090 49,938 50,564 47,879 307,094
ipt_1 56,674 42,167 47,458 55,440 52,499 53,179 52,046 53,980 53,978 59,832 59,037 57,158 643,448
ipt_2 47,320 33,835 36,371 47,093 44,029 46,436 46,290 47,214 47,231 54,798 54,265 53,454 558,336
ipt_3 33,752 23,765 24,242 33,219 31,766 35,736 36,112 36,764 36,310 46,059 46,188 45,907 429,820
ipt_4 missing missing 5,837 14,100 16,183 18,612 19,602 21,562 20,742 29,114 30,821 30,933 207,506
spontaneous_vaginal_delivery 48,839 46,788 47,810 45,582 46,445 44,261 46,979 47,426 49,240 51,005 50,124 48,080 572,579
deaths_from_post_abortions_complications 16 6 47 16 6 4 1 3 2 4 1 1 107
number_of_new_positives_put_on_arv missing missing missing missing missing missing missing 1 963 1,194 1,079 760 3,997
mothers_on_arv 903 818 843 761 989 1,303 1,243 1,308 39 missing missing 1 8,208
total_pnc_registrants 25,853 51,186 81,743 73,228 71,547 69,865 75,828 75,545 77,368 78,947 77,334 71,127 829,571
x1st_pnc_on_day_1_or_2_mother 25,853 51,149 55,671 52,103 52,155 51,422 54,993 56,029 61,650 69,096 69,400 64,911 664,432
x1st_pnc_on_day_3_7 missing 23 16,243 10,777 10,033 9,953 11,107 9,876 8,637 5,535 4,736 3,828 90,748
x1st_pnc_from_day_8_and_above missing 14 9,829 10,348 9,359 8,490 9,728 9,640 7,081 4,316 3,198 2,388 74,391
attendees_for_2nd_pnc 20,364 30,173 1,821 25 missing missing missing missing missing missing missing missing 52,383
x2nd_pnc_day_6_7_mothers missing missing missing missing missing missing missing missing missing missing missing missing missing
x3rd_pnc_at_6weeks_mothers missing missing missing missing missing missing missing missing missing missing missing missing missing
exclusive_breastfeeding_at_discharge 54,205 52,445 53,697 54,140 54,983 53,643 56,046 61,002 64,237 67,832 67,534 65,351 705,115
1 Sum

Association between health facility service delivery and maternal and neonatal outcomes.

Poisson regression modeling

  • GEE model
  • id = facility type to adjust for clustering by year
  • Correlation structure = exchangeable

For each model, we will consider the following as confounders

  1. Total ANC registrants (will ignore this as is the same as the numbers by age group)
  2. Mothers age at registration
  3. mothers_making_4th_anc_visit
  4. total_pnc_registrants (postpartum outcomes only)

Results below: No significant results found due to highlighted data issues: sample size.

Poisson regression: Marternal mortality

Basic model: Marternal mortality
Characteristic IRR1 95% CI1 p-value
antenatal_mother_at_registration_20_24 1.00007 0.99999, 1.00014 0.078
antenatal_mother_at_registration_25_29 1.00023 0.99991, 1.00055 0.2
antenatal_mother_at_registration_30_34 0.99987 0.99901, 1.00074 0.8
antenatal_mother_at_registration_35_39 0.99973 0.99927, 1.00020 0.3
mothers_making_4th_anc_visit 0.99998 0.99994, 1.00002 0.4
1 IRR = Incidence Rate Ratio, CI = Confidence Interval
IPT1: Marternal mortality
Characteristic IRR1 95% CI1 p-value
antenatal_mother_at_registration_20_24 1.00002 0.99993, 1.00012 0.6
antenatal_mother_at_registration_25_29 1.00029 0.99999, 1.00059 0.058
antenatal_mother_at_registration_30_34 0.99985 0.99903, 1.00066 0.7
antenatal_mother_at_registration_35_39 0.99968 0.99919, 1.00017 0.2
mothers_making_4th_anc_visit 0.99998 0.99994, 1.00002 0.3
ipt_1 1.00002 0.99998, 1.00006 0.3
1 IRR = Incidence Rate Ratio, CI = Confidence Interval
IPT2: Marternal mortality
Characteristic IRR1 95% CI1 p-value
antenatal_mother_at_registration_20_24 1.00007 1.00000, 1.00014 0.059
antenatal_mother_at_registration_25_29 1.00023 0.99995, 1.00051 0.11
antenatal_mother_at_registration_30_34 0.99988 0.99903, 1.00072 0.8
antenatal_mother_at_registration_35_39 0.99974 0.99916, 1.00032 0.4
mothers_making_4th_anc_visit 0.99998 0.99993, 1.00004 0.5
ipt_2 1.00000 0.99995, 1.00005 >0.9
1 IRR = Incidence Rate Ratio, CI = Confidence Interval
IPT3: Marternal mortality
Characteristic IRR1 95% CI1 p-value
antenatal_mother_at_registration_20_24 1.00007 0.99998, 1.00015 0.13
antenatal_mother_at_registration_25_29 1.00021 0.99993, 1.00048 0.14
antenatal_mother_at_registration_30_34 0.99989 0.99902, 1.00076 0.8
antenatal_mother_at_registration_35_39 0.99976 0.99915, 1.00038 0.5
mothers_making_4th_anc_visit 0.99999 0.99992, 1.00005 0.7
ipt_3 0.99999 0.99993, 1.00005 0.7
1 IRR = Incidence Rate Ratio, CI = Confidence Interval
spontaneous_vaginal_delivery: Marternal mortality
Characteristic IRR1 95% CI1 p-value
antenatal_mother_at_registration_20_24 1.00021 1.00000, 1.00041 0.052
antenatal_mother_at_registration_25_29 0.99996 0.99950, 1.00042 0.9
antenatal_mother_at_registration_30_34 0.99978 0.99863, 1.00093 0.7
antenatal_mother_at_registration_35_39 0.99966 0.99899, 1.00034 0.3
mothers_making_4th_anc_visit 0.99998 0.99994, 1.00002 0.3
spontaneous_vaginal_delivery 1.00009 1.00001, 1.00016 0.021
1 IRR = Incidence Rate Ratio, CI = Confidence Interval

Poisson regression: Still births

Basic model: Stillbirths
Characteristic IRR1 95% CI1 p-value
antenatal_mother_at_registration_20_24 1.00006 0.99999, 1.00012 0.076
antenatal_mother_at_registration_25_29 1.00012 1.00009, 1.00015 <0.001
antenatal_mother_at_registration_30_34 0.99976 0.99962, 0.99991 0.002
antenatal_mother_at_registration_35_39 1.00001 0.99987, 1.00015 0.9
mothers_making_4th_anc_visit 1.00003 1.00002, 1.00004 <0.001
1 IRR = Incidence Rate Ratio, CI = Confidence Interval
IPT1: Stillbirths
Characteristic IRR1 95% CI1 p-value
antenatal_mother_at_registration_20_24 1.00006 0.99997, 1.00015 0.2
antenatal_mother_at_registration_25_29 1.00012 1.00010, 1.00013 <0.001
antenatal_mother_at_registration_30_34 0.99977 0.99962, 0.99991 0.001
antenatal_mother_at_registration_35_39 1.00001 0.99985, 1.00017 0.9
mothers_making_4th_anc_visit 1.00003 1.00002, 1.00004 <0.001
ipt_1 1.00000 0.99998, 1.00002 >0.9
1 IRR = Incidence Rate Ratio, CI = Confidence Interval
IPT2: Stillbirths
Characteristic IRR1 95% CI1 p-value
antenatal_mother_at_registration_20_24 1.00008 1.00000, 1.00015 0.041
antenatal_mother_at_registration_25_29 1.00008 1.00005, 1.00011 <0.001
antenatal_mother_at_registration_30_34 0.99979 0.99965, 0.99993 0.003
antenatal_mother_at_registration_35_39 1.00005 0.99983, 1.00027 0.6
mothers_making_4th_anc_visit 1.00003 1.00001, 1.00005 <0.001
ipt_2 0.99999 0.99995, 1.00003 0.5
1 IRR = Incidence Rate Ratio, CI = Confidence Interval
IPT3: Stillbirths
Characteristic IRR1 95% CI1 p-value
antenatal_mother_at_registration_20_24 1.00007 1.00003, 1.00011 0.001
antenatal_mother_at_registration_25_29 1.00006 1.00001, 1.00010 0.015
antenatal_mother_at_registration_30_34 0.99980 0.99966, 0.99993 0.003
antenatal_mother_at_registration_35_39 1.00009 0.99982, 1.00036 0.5
mothers_making_4th_anc_visit 1.00004 1.00001, 1.00007 0.021
ipt_3 0.99997 0.99991, 1.00004 0.4
1 IRR = Incidence Rate Ratio, CI = Confidence Interval
spontaneous_vaginal_delivery: Stillbirths
Characteristic IRR1 95% CI1 p-value
antenatal_mother_at_registration_20_24 1.00008 0.99999, 1.00017 0.088
antenatal_mother_at_registration_25_29 1.00007 0.99999, 1.00015 0.084
antenatal_mother_at_registration_30_34 0.99975 0.99957, 0.99994 0.009
antenatal_mother_at_registration_35_39 0.99998 0.99978, 1.00018 0.9
mothers_making_4th_anc_visit 1.00003 1.00002, 1.00004 <0.001
spontaneous_vaginal_delivery 1.00002 0.99998, 1.00006 0.4
1 IRR = Incidence Rate Ratio, CI = Confidence Interval

Poisson regression: Neonatal mortality

I have not run this analysis. There is nothing we are going to find here. Need data from all health facilities to check what is happening. The increase in NM should be consistent across health facilities if not an error. This trend might be driven by a few Hospitals with data issues and this can only be verified with health facility level data.