Research Artcle MedPub Journals http://www.medpub.com/ Journal of Healthcare Communcatons ISSN 2472-1654 DOI: 10.4172/2472-1654.10002 Relablty of usng Standardzed Mortalty Rato n Estmaton of Relatve Rsk of Under-Fve Mortalty n Uganda; an Emprcal Analyss Asmwe JB School of Statstcs and Plannng, Makerere Unversty, P.O. Box 7062, Kampala, Uganda Correspondng Author: Asmwe JB Abstract In Uganda, usng survey data, estmates of under-fve mortalty have only been avalable at natonal and regonal levels. However, estmaton of under-fve mortalty can be made for dstrcts usng small area estmaton technques. The smpler way s to use the Standardzed Mortalty Rato (SMR). Lterature has shown that use of SMR s subject to unrelable results but no emprcal study has verfed that ths s so. The author used Uganda Demographc and Health Survey data of 1995, 2001 and 2006 n the nvestgatons to explore emprcally how relable s the use of SMR n estmaton of relatve rsk of under-5 mortalty. The author appled the coeffcent of varaton to measure relablty of the SMR estmates. Utlzaton of the tradtonal SMR could potentally be assocated wth very hgh coeffcent of varatons. The author recommends that before utlzaton of SMR, there s need to explore relablty of the results usng coeffcent of varaton. Key Words: Coeffcent of varaton; Dstrct; SMR asmweajb@gmal.com School of Statstcs and Plannng, Makerere Unversty, P.O. Box 7062, Kampala, Uganda Tel: +256772428489 Ctaton: Olowsk P, Mchelo C. Dfferental Burden and Determnants of Tobacco Smokng: Populaton-Based Observatons from the Zamba Demographc and Health Survey (2002 and 2007). J Healthc Commun. 2016, 1:1. Receved: Oct 20, ; Accepted: Nov 16, ; Publshed: Nov 23, Introducton Standardzed Mortalty Rato s one of the basc methodologes used n small area estmaton technque and s smply a rato of observed to the expected deaths and could be used to provde estmates of local area such as dstrct or sub-county whose estmates at that level may not be derved by drect estmaton from the avalable survey data. Small area estmaton s statstcal technques nvolvng the estmaton for small sub-populatons, generally used when the sub-populaton of nterest s ncluded n a larger survey. For example a natonal survey may derve estmates for regonal cluster statstcs but not for dstrct level but usng small area estmaton ndcators for the latter could be estmated. Small area estmates n general may be useful for government agences to allocate resources or dentfy hazardous areas related to hgh under-fve mortalty so that approprate acton may be taken [1-3]. Mappng mortalty and dsease rates to dsplay geographc varablty s an ncreasngly common epdemologcal tool and falls under a broad subject of small area estmaton. Understandng spatal clusterng of chldhood or/and under-fve mortalty can provde a gude n targetng nterventons n a more strategc approach to the populaton where mortalty s hghest and the nterventons are most lkely to make an mpact [4]. Natonal surveys are wdely used to provde estmates for the entre populaton parameters of nterest but also for subpopulatons (domans) such as regons, rural or urban, sex and age groups. However, such subpopulatons are generally too large to provde a sense of partcular lower level localtes (small areas) lke dstrct or countes or sub-countes where the actual problem can easly be located. Interventon can easly be accomplshed when a small localty has been dentfed wth a partcular problem. Small area estmaton provdes a soluton to usng survey data to furnsh estmates at such lower localtes. The dea s that small area estmaton technques n partcular borrow strength by usng values of the varable of nterest, y, from related areas to ncrease effectve sample sze. The value of, y, s by tself small to provde a relable drect estmate for a partcular localty. For example, the number of under-fve deaths, Copyrght MedPub Ths artcle s avalable n: http://healthcare-communcatons.medpub.com/archve.php 1
y, derved from a natonal survey data may be too small to provde estmate of under-fve mortalty for a partcular dstrct. However, usng small area estmaton, the value of, y, can help derve a relable estmate for relatve rsk of under-fve mortalty for the dstrct. Relatve Rsk s the rato of the ncdence of dsease n the exposed populaton to the ncdence n the non-exposed populaton [5]. It s a rato of two probabltes. In utlzng small area estmaton technques, the values, y, are brought nto the estmaton process through a model that provdes a lnk to the related areas. A study by Asmwe et al., [6] show that use of Posson-gamma and log-normal models offer relable wth less nose n the estmates for small area under-5 mortalty data though these methodologes appear to be more complex. The smplest form s to use SMR whch s a rato of the observed value, y, to the expected deaths to provde an estmate of the relatve rsk of under-fve mortalty for a gven localty. However, lterature has shown that use of SMR s subject to unrelable results but no emprcal study has verfed that ths s so. Materals and Method The study utlzed data obtaned from Uganda Demographc and Health Surveys of 1995, 2001 and 2006. Ths secton provdes dscusson on the three data sets that were used n the study; ther sources and weakness. A summary of the key characterstcs of the three UDHS surveys of 1995, 2001 and 2006 are gven n Table 1. The UDHS survey of 1995 covered a total of 37 dstrcts and due to armed conflct; the dstrct of Ktgum located n the northern part of the country was not covered. By the tme of the survey, Uganda had a total of 38 dstrcts. A sample of 303 Prmary Samplng Unts (PSUs) consstng of Enumeraton Areas (EAs) were selected from a samplng frame of the 1991 Populaton and Housng Census and covered a total of 7,070 women n the reproductve age group of 15 49 years. An EA n most cases s equvalent to a vllage but where the sze of the vllage s bg two or more EAs are created. The survey also obtaned data from a total of 7,550 households and 1,996 men n a reproductve age group of 15-54 years. The country was clustered nto four regons consstng of Central, Eastern, Northern and Western. To permt calculaton of contraceptve prevalence rates under a USAID-funded project called DISH (Delvery of Improved Servces for Health) a sample desgn allowed for over samplng of households n the nne dstrcts. These dstrcts were Kasese, Mbarara, Masaka, Raka, Luwero, Masnd, Jnja, Kamul and Kampala. Over samplng allowed for a relable sample sze that would purposvely allow derve estmates for such areas or dstrcts. The UDHS data of 2001 covered a total of 34 dstrcts and agan due to armed conflct; the dstrcts of Kasese and Bundbugyo n Western Regon as well as Gulu and Ktgum n the Northern Table 1 Summary of the key characterstcs of the 1995, 2001 and 2006 UDHS data sets. Year of Survey Data Number of Dstrcts 1995 37 7,550 2001 34 7,885 2006 56 8,870 Number of Households Sampled regon were excluded from the survey. A sample of 298 PSUs consstng of EAs were selected from a samplng frame of the 1991 Populaton and Housng Census and covered a total of 7,246 women n the reproductve age group of 15-49 years. The survey equally obtaned data from a total of 7,885 households and 1,962 men n a reproductve age group of 15-54 years. The country was also clustered nto four regons consstng of Central, Eastern, Northern and Western. To permt calculaton of contraceptve prevalence rates under DISH project the nne dstrcts were agan over sampled. Over samplng of EAs were also carred out for the dstrcts (Kabale, Ksoro and Rukungr) under the project called CREHP (Communty Reproductve Health Project). As compared to the UDHS of 1995 and 2001, the survey of 2006 covered all the 56 dstrcts of the country provdng a better estmate of SMR. Although the number of dstrcts n the country have contnued to ncrease to currently more than 110, by the tme of the 2006 survey there were only 56. A total sample of 321 prmary samplng unts (PSU) consstng of enumeraton areas (EAs) were selected from a samplng frame of the clusters sampled n the 2005-2006 Uganda Natonal Household Survey and an addtonal 47 EAs were over sampled from the North Eastern Regon (Kotdo, Moroto and Nakaprprt) and IDP camps n the dstrcts of Gulu, Ktgum, Lra and Pader [7]. The over sampled areas were amed at obtanng specfc baselne ndcators due to armed conflct that ravaged the regon for over 20 years pror to the survey. The country was clustered nto nne regons compared to the four covered n the pror surveys. The nne regons ncluded; Central 1, Central 2, Kampala, Eastern, East Central, North, West Nle, Western and South Western (Fgure 1). In general, a total of 8,531 women n the reproductve age group of 15 49 years were ntervewed. The survey equally obtaned data from a total of 8,870 households and 2,503 men n the reproductve age group of 15-54 years. In each of the UDHS surveys of 1995, 2001 and 2006, data on the number of chldren dead reported by women n a reproductve age group of 15-49 years were collected. Data was accessed wth permsson from Demographc and Health Survey webste. Ths research work used under-fve mortalty rather than other mortalty measures lke nfant or chld mortalty due to the fact that sgnfcantly larger and more samples are obtanable. Addtonally, the ndcator s n lne wth the MDGs target to reduce under-fve mortalty by two-thrds between 1990 and. The data used n the estmaton of the under-fve mortalty rates were collected on the brth hstory of women aged 15-49 years. For chldren who had ded, the women were asked to provde the age at death. The data used for computaton of under-fve mortalty s susceptble to some errors. Frstly, only survvng women aged 15-49 years were ntervewed; therefore, no data are avalable for chldren of women who had ded. Another possble error n data collecton s underreportng of events (brths and deaths), especally n cases where deaths occur early n nfancy. Attempts to address under reportng of age at deaths were done by recordng days f the death took place wthn one month after brth, n months f the chld ded wthn 24 months, and n years f the chld was two years or older [8]. The author used the Standardzed Mortalty Rato (SMR) and s 2 Ths artcle s avalable n: http://healthcare-communcatons.medpub.com/archve.php
Fgure 1 defned as; y e θ = (1) Where y and e denote the number of deaths and the expected number of deaths respectvely from the dsease durng the study perod. Generally the expected number of cases e s assumed known (Baley, 2001). The standard devaton (s.d) for the relatve rsk under the SMR s gven as [9]. 1 sd. = (2) y where y refer to the number of observed cases of under-fve deaths n a gven dstrct,. The standard error (s.e) for the relatve rsk under the SMR s gven as; e s.e y = (3) where e refers to the expected number of deaths n a gven dstrct. Coeffcent of varaton s defned as; sd. cv = *100 (4) SMR Results Map of Uganda showng clusters used n UDHS 2006. Usng the 1995 UDHS data, SMR had lower CVs (<100%). Ths may largely be attrbuted to the fact that the number of dstrcts was stll few by 1995 and the observed counts were farly substantal. By 1995, there were a total of 38 dstrcts although the demographc and health survey covered 37 due to armed conflct n one of the dstrct. Compared to 2006 where we had a total of 56 dstrcts and even f the sample sze had slghtly ncreased, SMR showed a more relable and stable estmates wth fewer dstrcts for ether 1995 or 2001. The hghest CV was 59.8% obtaned for Kabale dstrct. Under Lcense of Creatve Commons Attrbuton 3.0 Lcense SMR results usng the 2001 UDHS data showed very hgh varablty (>100%) n three dstrcts of Kapchorwa, Kotdo and Homa as shown n Table 2. Overall, other dstrct s coeffcent of varaton was relatvely low ndcatng low level of nose n the SMR computatons. Despte hgh CVs n the three dstrcts, overall the other dstrcts showed lower CVs and agan ths mght be attrbuted to the fact that the numbers of dstrcts were stll few (34) and subsequently a large number of observatons per dstrct to reduce the nose. Results obtaned from the standard devaton of SMR show hgh values for the dstrcts of Adjuman, Kaberamado, Ksoro, Mayuge, Moyo and Yumbe. The coeffcent of varaton (CV) for these dstrcts were relatvely very hgh (>100%) as shown n Table 2 depctng unrelablty n utlzaton of SMR to estmate relatve rsk of under-fve mortalty. The nose from SMR results can be attrbuted to the fact that more dstrcts (56) were ntroduced that reduced the sample sze per dstrct. Dscusson Coeffcent of varaton (CV) usng SMR for the 1995 UDHS data showed lttle varablty or smply the standard devatons were small. In all the cases for the 37 dstrcts none of them exceeded 60% when usng CV. These results further show that when few dstrcts are nvolved n estmaton of SMR and when substantal data ponts are provded, SMR estmates appear to be a good estmate of relatve rsk of under-fve mortalty. SMR results usng the 2001 UDHS data showed very hgh varablty (>100%) n three dstrcts of Kapchorwa, Kotdo and Homa as shown n Table 2. Overall, other dstrct s coeffcent of varaton was relatvely low ndcatng low level of nose n the SMR computatons. Agan ths may largely be attrbuted to the fact that the number of dstrcts was stll few by 2001 and the observed count (y ) was farly substantal. SMR results usng the 2006 UDHS data showed very hgh CV for the dstrcts of Adjuman, Kaberamado, Ksoro, Mayuge, Moyo and Yumbe. The coeffcent of varaton (CV) for these dstrcts were relatvely very hgh (>100%) depctng unrelablty n utlzaton of SMR to estmate relatve rsk of under-fve mortalty. Concluson There were 37, 34 and then 56 dstrcts n the UDHS data of 1995, 2001 and 2006 respectvely. Snce there were fewer dstrcts n the UDHS of 1995 and 2001, there were more observatons per dstrcts for these perods. More observatons allowed for less volatlty n SMR measure compared to the UDHS data of 2006. Wth UDHS data of 2006 less observaton per dstrcts were gvng rase to ncrease nose n the SMR results. Recommendaton The author therefore recommend that before utlzaton of SMR, there s need to explore emprcally the relablty of the results usng smple technques such as coeffcent of varaton. We also recommend use of alternatve Bayesan approaches lke Besang, York, Molle, Posson-gamma or the Log-normal models to smoothen the estmates. 3
Table 2 Varablty arsng from use of SMR as shown by Coeffcent of Varaton (CV%) usng 2006, 2001 and 1995 UDHS data. 2006 2001 1995 Standard Standard Standard No. Dstrct CV% for CV% for SMR Devaton SMR Devaton SMR Devaton SMR SMR (SMR) (SMR) (SMR) CV% for SMR 1 Adjuman 0.1 0.316 248.9 - - - - - - 2 Apac 0.6 0.076 13.1 0.59 0.090 14.5 0.46 0.107 23.2 3 Arua 0.8 0.061 7.3 0.57 0.080 14.2 0.67 0.075 11.3 4 Bugr 0.5 0.139 31.0 - - - - - - 5 Bundbugyo 0.9 0.132 15.5 - - - 1.62 0.136 8.4 6 Busheny 0.5 0.080 14.9 0.63 0.080 12.9 0.63 0.087 13.7 7 Busa 0.6 0.169 30.6 - - - - - - 8 Gulu 1.0 0.069 6.9 - - - 0.63 0.106 16.7 9 Homa 0.5 0.132 25.4 0.17 0.230 135.8 0.66 0.164 25.0 10 Iganga 1.0 0.072 7.5 1.00 0.070 7.2 0.60 0.068 11.4 11 Jnja 0.2 0.196 82.3 0.50 0.140 27.4 0.68 0.115 16.8 12 Kabale 0.4 0.124 34.8 0.57 0.110 19.1 0.29 0.171 59.8 13 Kabarole 0.2 0.196 85.8 1.30 0.080 6.2 0.63 0.086 13.7 14 Kaberamado 0.3 0.333 124.3 - - - - - - 15 Kalangala 0.6 0.378 65.7 2.06 0.240 11.5 4.16 0.218 5.2 16 Kampala 0.5 0.089 17.2 0.35 0.100 27.3 0.62 0.082 13.3 17 Kamul 1.0 0.072 7.5 0.99 0.070 7.3 1.19 0.067 5.6 18 Kamwenge 0.9 0.116 13.2 - - - - - - 19 Kanungu 0.4 0.186 52.3 - - - - - - 20 Kapchorwa 0.4 0.229 58.6 0.06 0.580 1003.0 0.45 0.224 50.1 21 Kasese 0.4 0.123 31.1 - - - 0.35 0.171 49.3 22 Katakw 0.4 0.192 54.2 - - - - - - 23 Kayunga 0.3 0.192 59.4 - - - - - - 24 Kbaale 0.6 0.114 19.2 0.53 0.120 22.3 1.11 0.120 10.8 25 Kboga 0.5 0.177 36.0 0.90 0.140 15.4 1.24 0.136 11.0 26 Ksoro 0.1 0.354 388.1 1.56 0.090 6.0 0.92 0.143 15.5 27 Ktgum 0.9 0.096 11.0 - - - - - - 28 Kotdo 1.1 0.064 6.0 0.11 0.210 189.7 1.22 0.186 15.2 29 Kum 0.3 0.171 50.0 0.65 0.120 18.6 1.04 0.103 9.9 30 Kyenjojo 0.6 0.121 21.4 - - - - - - 31 Lra 0.5 0.082 18.2 0.49 0.090 18.6 0.70 0.083 11.9 32 Luwero 0.4 0.143 39.6 0.49 0.130 26.5 0.61 0.108 17.6 33 Masaka 0.5 0.085 16.4 0.98 0.070 7.5 0.88 0.066 7.5 34 Masnd 0.5 0.113 21.0 0.20 0.180 91.9 0.38 0.189 50.2 35 Mayuge 0.1 0.354 404.5 - - - - - - 36 Mbale 0.9 0.077 8.5 0.74 0.080 11.3 0.61 0.078 12.8 37 Mbarara 0.4 0.075 18.4 0.62 0.070 10.8 1.11 0.058 5.2 38 Moroto 1.5 0.095 6.2 0.80 0.140 17.7 0.46 0.174 38.1 39 Moyo 0.2 0.277 161.6 1.04 0.120 11.7 0.58 0.154 26.7 40 Mpg 0.3 0.156 54.3 1.08 0.100 8.8 0.54 0.081 15.0 41 Mubende 0.9 0.075 8.3 1.02 0.080 7.3 0.94 0.083 8.8 42 Mukono 0.5 0.098 21.3 0.55 0.100 17.3 0.47 0.092 19.7 43 Nakaprprt 1.0 0.128 12.4 - - - - - - 44 Nakasongola 0.6 0.209 32.8 - - - - - - 45 Nebb 0.5 0.107 20.8 0.66 0.100 15.6 0.62 0.035 5.7 46 Ntungamo 0.3 0.139 40.4 - - - - - - 47 Pader 1.0 0.085 8.8 - - - - - - 48 Pallsa 0.5 0.124 25.3 0.50 0.120 23.4 0.70 0.102 14.6 49 Raka 0.7 0.095 14.1 0.76 0.110 13.8 0.70 0.110 15.7 50 Rukungr 0.3 0.165 50.2 1.68 0.080 4.8 0.44 0.143 32.5 51 Sembabule 1.1 0.123 11.7 - - - - - - 4 Ths artcle s avalable n: http://healthcare-communcatons.medpub.com/archve.php
52 Sronko 0.4 0.196 54.5 - - - - - - 53 Sorot 0.4 0.171 47.5 0.70 0.120 17.0 0.64 0.097 15.1 54 Tororo 0.6 0.113 19.8 0.62 0.110 17.0 0.64 0.086 13.5 55 Wakso 0.2 0.126 63.5 - - - - - - 56 Yumbe 0.2 0.250 153.2 - - - - - - The symbol - ndcates that the dstrct was not yet created Acknowledgement I am grateful to Unted States Agency for Internatonal Development (USAID) for provdng onlne data sets that I used n ths study. Under Lcense of Creatve Commons Attrbuton 3.0 Lcense 5
References 1 Wakefeld J (2007) Dsease Mappng and Spatal Regresson wth Count Data. Bostatstcs 8: 158-183. 2 Meza JL (2002) Emprcal Bayes estmaton smoothng of relatve rsks n dsease mappng. Journal of Statstcal Plannng and Inference 112: 43 62. 3 Lawson AB, Bgger AB, Boehnng D, Lesaffre E, Vel JF, et al. (2000) Dsease Mappng Models: An Emprcal Evaluaton. Stat Med 19: 2217-2241. 4 Lutamb AM, Alexander M, Jensen C, Mahutanga C, Nathan R (2010) Spatal-temporal clusters n Ifakara HDSS n South-eastern Tanzana. Glob Health Acton 3. 5 Lawson AB, Wllams FLR (2001) An ntroducton Gude to Dsease Mappng. UK: John Wley & Sons Ltd., West Sussex. 6 Asmwe JB, Jehopo P, Atuhare LK, Mbonye AK (2011) Examnng small area estmaton technques for publc health nterventon: Lessons from applcaton to under-fve mortalty data n Uganda. J Publc Health Polcy 32: 1-15. 7 UBOS (2005) Uganda Natonal Household Survey 2005/2006-Report on the Soco-Economc Survey. Uganda Bureau of Statstcs (UBOS), Entebbe, Uganda. 8 UBOS and ORC Macro (2007) Uganda Demographc and Health Survey 2006. Uganda Bureau of Statstcs, Kampala, Uganda. 9 Soe MM, Sullvan KM (2006) Standardzed Mortalty Rato and Confdence Interval. 6 Ths artcle s avalable n: http://healthcare-communcatons.medpub.com/archve.php