The TCDI Kenya website was developed by Development Gateway in partnership with the University of Cape Town. The website is intended to respond to the needs of government, civil society and academic stakeholders by providing access to demand-driven, high-quality data on national tobacco prevalence, products, and policies. The design, functionality and content for the website are led by the Kenyan tobacco control community’s needs. This document outlines our overall website methodology.

TCDI Kenya used the 2014 Kenya Global Adult Tobacco Survey (GATS) data to calculate tobacco use prevalence for different products and sub-groups of the population.  The analysis was conducted using the statistical software, Stata. The Kenya GATS data is available for download here, and the Stata code here.

To estimate the trend in adult male tobacco use and tobacco use by region, the 2014 Kenya Demographic and Health Data (KDHS) was used. To estimate the population size, we used UNDP mid-year population estimates for 2014. The DHS data is available for download here, and the Stata code here.       

The 2014 GATS is a national household-based sample survey of adults aged 15 years and older who are members of a selected household.  It defines a household as a person or group of persons who normally reside together in the same compound under one or several roofs, are answerable to the same head, and share a common cooking arrangement (Kenya Ministry of Health and Kenya National Bureau of Statistics, 2014; page 13). One individual was randomly selected from all eligible listed household members. 

The Kenya GATS 2014 is a multistage probability household sample survey covering all 47 counties to provide national estimates for tobacco indicators for rural and urban areas, and to allow separate national estimates for males and females. The Kenya GATS 2014 adopted a three-stage cluster sample design, with the first stage involving the selection of sample points (‘clusters’) while second and third stages involved selection of households and eligible individuals, respectively. A representative sample of 5,376 households was drawn with a target of interviewing one eligible adult, aged at least 15 years, within the sampled households.

The survey used the fifth National Sample Surveys and Evaluation Programme (NASSEP V) master sample frame that was developed and maintained by  the Kenya National Bureau of Statistics (KNBS). The frame was developed using the enumeration areas generated from the 2009 Kenya Population and Housing Census and has 5,360 clusters split into four equal sub-samples. The sample size for 2014 Kenya GATS was 5,376 households selected from a total of 192 clusters, 102 in urban and 90 in rural, with a uniform sample of 28 households per cluster. 

The GATS provides data on demographics, tobacco smoking, smokeless tobacco use, cessation, second hand smoke, among other variables.

To estimate the smoking prevalence, we used the daily smokers and occasional smokers, defined as the use of tobacco within the past 30 days, preceding the survey, either daily or occasionally. Daily tobacco use is defined as consumption of at least one tobacco product every day or nearly every day over a period of a month. The number of smokers are divided by the total sample and multiplied by weights to estimate the smoking prevalence rate as shown below.

Where i is an individual in the survey,
x_i is the population group of interest, e.g males
s_i  are the number of current smokers in the population group of interest e.g. male smokers, and
w_i are the survey weights associated with those individuals.

The TCDI team calculated prevalence rates for the following groups: 

• Six age categories were identified for the smoking prevalence by age analysis: 15 to 24 years, 25 to 34 years, 35 to 44 years, 45 to 54 years, 55 to 64 years, and 65+ years.  An additional age category was included for those aged 13 to 15 years for 2001, 2007 and 2013 smoking prevalence estimates, based on the Kenya Global Youth Tobacco Survey (GYTS). The Kenya GYTS data can be found here and the Stata code for the prevalence estimates can be found here.
• For tobacco smoking prevalence among primary school going children, data from the National Agency for the Campaign Against Drug Abuse (NACADA) was used. The prevalence was estimated for school going children aged 8 to 13 years old, 14 to 16 years old, and 17 to 20 years old. The NACADA report can be found here.
• For the level of education analysis, respondents were asked about the highest level of education completed. Responses were divided into no education, primary education incomplete, primary education completed  (​​which includes primary school completed and less than secondary school completed), and secondary education or higher (which includes secondary school completed, tertiary college completed, university completed, and post graduate degree completed).
• For tobacco products,  any tobacco, manufactured cigarettes, hand rolled cigarettes, smokeless tobacco and snuff were included in the analysis.  Smokeless tobacco is defined as products wholly or partly made of tobacco and do not need to be ignited for consumption. Common smokeless tobacco products found in Kenya include chewing tobacco, snuff, kuber and betel quid. These tobacco products are either found un-packaged (wrapped in various materials such as banana leaves) or branded packets. Smoked tobacco is defined as products wholly or partly made of tobacco and requires to be ignited to enable consumption. The smoked products assessed in the GATS include manufactured cigarettes, hand rolled cigarettes, pipe (kiko), cigars and shisha (Kenya Ministry of Health and Kenya National Bureau of Statistics, 2014; page 152).
• For location, households were categorised into urban and rural.

Why do we use the GATS, GYTS and DHS data?

The International Tobacco Control Survey Wave 2 is the most recent survey on tobacco use behaviour. It is more focused on measuring the psychosocial and behavioural impact of key tobacco control policies rather than prevalence.  The TCDI team sought to use open access, nationally representative data for the tobacco prevalence estimates for Kenya. The Kenya STEPS 2015 data provides data on tobacco prevalence for a variety of  tobacco products, but limits its estimates to persons aged 18 to 69 years. The TCDI team used the Kenya GATS 2014 data for estimates of tobacco prevalence as the GATS provides estimates on tobacco use for a variety of tobacco products and includes a wide range of ages from 15 years to 65 years and above. However, The GATS does not allow for disaggregating tobacco prevalence estimates by region. For the estimates on tobacco use by region, the TCDI team uses the Kenya DHS 2014 data, which provides estimates on tobacco use for men aged 15 to 54 years and women aged 15 to 49 years. To estimate smoking prevalence among the youth, the Kenya GYTS (youth aged 13-15 years) and NACADA (school children aged 8 -20 years) data were used.

For the tobacco taxation and illicit trade module, the TCDI team drew on secondary sources,  and conducted data analysis. The team undertook their initial data search and literature review between November 2021 and March 2022 and the data analysis took place from January 2022 to May 2022. Secondary data searches occurred across various platforms, including data repositories, academic journals, and online libraries. Additionally, the team reviewed resources from academic institutions such  as Tobacco Tactics from the University of Bath, and government departments  such as the  Kenya Revenue Authority (KRA) , and the Kenya National Bureau of Statistics (KNBS). The team also reviewed resources from trusted organizations such as the World Bank Group. The TCDI team consulted with the Research Unit on the Economics of Excisable Products (REEP) and subject matter experts, who advised on data sources and reviewed each page before publication.

For the data analysis conducted by the TCDI team, the following data and methods were used:

1. Cigarette affordability, 1990-2018

The team used data from the Kenya National Bureau of Statistics, various years for consumer price indices, and Sportsman cigarette prices (Statistical Abstract, 1995-2019), and data from the World Bank for the GDP per capita.  The affordability index, was estimated as the ratio of the real (inflation-adjusted) GDP per capita to the real cost of 100 packs of Sportsman cigarettes  (see equation 1).

Where t refers to the year,  t=1990, 1991, 1992,…, 2018

A lower reading on the affordability index suggests that cigarettes are less affordable while a higher reading suggests that  cigarettes are more affordable, i.e. cost less.

2. Cigarette consumption and tobacco excise tax revenue projections to 2025

To estimate  cigarette consumption by 2025 if the tobacco tax policies in Kenya remain constant, the TCDI team used the Tobacco Excise Tax Simulation Model (TETSiM).

The Tobacco Excise Tax Simulation Model (TETSiM) is a simulation tool that has been developed by the Research Unit on the Economics of Excisable Products at the University of Cape Town. The TETSiM model quantifies the likely impact of a change in the excise tax structure and/or the level of the excise tax on a number of variables, including cigarette consumption and excise revenue collected by the government. Because the model is programmed in Excel,

it can be customized to the tax system for any individual country.

The team designed 2 scenarios for Kenya:

While an over-shift may not be the response from the tobacco industry, we assumed an over-shift because in cases of highly concentrated markets, like in Kenya, industries are likely to over-shift the tax.

The following assumptions were made for all scenarios modeled

1. Product segments: Six brand segments, including imported, premium, popular, discount, unfiltered, and illicit
2. Illicit cigarette market: 5% of the total cigarette market based on the estimates by the Kenya Revenue Authority
   3
3. Price elasticity of demand: Range -0.6 to -0.8, with absolute elasticity being greater for lower income groups, and smaller for higher income groups.
   3
4. Cross-price elasticity of demand: Range 0 to 0.2, with 0 for unfiltered brand category, 0.2 for popular brands, and 0.1 for discount and illicit brands.
5. Income elasticity of demand: Range  from 0.3 (illicit brands) to 0.5 (premium brands). The income elasticity was assumed to be 0.4 for popular, discount and unfiltered brands.  

Note:

• The price elasticity of demand measures the percentage change in the demand of a product resulting from a 1% change in the retail price of that product. In the case of cigarettes, price elasticity is negative, because increases in price will result in a reduction in demand. In low- and middle-income countries (LMICs), the price elasticity of demand for cigarettes has been estimated at between -0.2 and -0.8, with many estimates clustered at around -0.5. Thus, for LMICs, a 10% increase in the retail price of cigarettes is expected to reduce consumption by between 2% and 8%.
• The cross-price elasticity of demand measures by what percentage the demand of one brand segment changes in response to a 1% increase in the price a more expensive brand of cigarettes. Since cross-price elasticity reflects the gains to the cheaper segment (from the more expensive segment) it is a positive number. In other words, it captures how consumers substitute to cheaper products when prices increase. This also allows for movement between the legal and illegal markets, as some consumers switch from legal to (presumably cheaper) illegal cigarettes. ^[3] This range of elasticity is in line with international estimates for LMICs.
• The income elasticity of demand measures by what percentage the demand changes in response to a 1% increase in income. As income rises, the demand for normal  goods increases.
  4
  5
   Thus, the income elasticity for normal goods is positive, implying that the demand rises with income. Goods with income elasticities above one are perceived as luxury or superior goods. While those with income elasticities below one are perceived as neccessities.
  6
    Most estimates of income elasticity of demand for tobacco products lie between 0 and 1. An income elasticity of demand of 0.5 means that an increase in income of 10% will result in tobacco consumption rising by 5%.
  7

Data from the International Monetary Fund,  Kenya Revenue Authority, the United Nations data repository World Bank , and the World Health Organization were used for the simulation of all 4 models.

The TCDI team utilized primary and secondary data collected and analyzed by the Tobacco Control Data Initiative (TCDI) in 2022 to determine  the morbidity and mortality levels stemming from tobacco use in Kenya and the economic costs thereof.

Mortality

Secondary data on mortality from the Civil Registration and Vital Statistics Unit within the Kenyan Ministry of Health was analyzed by the TCDI team. Specifically, the Vital Statistics Unit database collects all-cause mortality data from hospitals in 47 counties and the four national reference hospitals in Kenya. This data, which dated between 2012 and 2021. Cause of death was based on the death certificates completed by clinicians in hospitals and entered into the Kenya Health Information System.

The data on mortality attributable to tobacco use was abstracted for five tobacco-related illnesses (respiratory diseases, diabetes mellitus, malignant cancers, tuberculosis, and cardiovascular diseases). This abstraction was based on the sequence of events leading to death indicated on the death certificates and coded based on the 10th International Disease Classifications (ICD-10) rules and guidelines.

Based on accurate cause of death reports, the number of cases with ill-defined causes of death (due to inaccurate diagnosis) was compared with the tobacco-related death codes recorded between 2012 and 2021. The rate of ill-defined cause of death ranged from 7% to 5%. Mortality data based on the date of occurrence from all health facilities within Kenya were included in this analysis, while all data from lay reporting and not coded according to ICD 10 guideline and rules were excluded.

This secondary data was used to model the deaths attributable to tobacco smoking in Kenya between 2012 and 2021. The study used a prevalence-based analysis in the cohort studies model to calculate adjusted smoke-attributable mortality (SAM) rates for a person 35 years or older. This analysis used the population attributable fraction (PAF) associated with tobacco smoking to calculate excess-mortality prediction.

The PAF was produced using the prevalence of tobacco smoking estimates from the STEPs Survey 2015 and the Global Adult Tobacco Survey (GATS) 2014. GATS was used to estimate the age and gender-specific prevalence of tobacco smoking in Kenya. GATS estimated that 15.1% of men, 0.8% of women, and 7.8% of people overall (1.7 million adults) were current tobacco smokers. This data was combined with data from the STEPs survey to compute an average for each age category. The STEP survey estimated that 19.7% of men, 0.9% of women, and 10.1% of people overall smoked tobacco. The combined data produced the mid-decade estimates for the prevalence of tobacco smoking in Kenya, as shown in Table 1 below:

	Current	Former	Never
Sex
Male	24.05	17.25	58.7
Female	1.35	2.35	96.3
Age
35-64	1.45	1.9	96.65
65+	0.65	6.1	93.25

The relative risks were derived from available literature and drawn from the American Cancer Society Cancer Prevention Study II, as shown below in Table 2.

Table 2: Relative Risks of Various TRI

	Relative Risk
Disease	Current Smoker Female	Former smoker Female	Current Smoker Male	Former Smoker Male
1. Malignant cancers (Cancer)
Esophagus	7.8	2.8	6.8	4.5
Kidney and renal pelvis	1.3	1.1	2.7	1.7
Larynx	13	5.2	14.6	6.3
Lips, oral cavity, pharynx	5.1	2.3	10.9	3.4
The neck of the uterus	1.3	1.1
Pancreas	2.3	1.6	2.3	1.2
Stomach	1.4	1.3	2	1.5
Trachea, lungs, bronchi	12.7	4.5	23.3	8.7
Urinary bladder	2.2	1.9	3.3	2.1
Acute myeloid leukemia	1.1	1.4	1.9	1.3
2. Cardiovascular diseases (CVD)
Cerebrovascular disease {35-64 years}	4	1.3	3.3	1
Cerebrovascular disease {≥65 years}	1.5	1	1.6	1
Ischemic heart disease (IHD) {35-64 years}	3.1	1.3	2.8	1.6
Ischemic heart disease (IHD) {≥65 years}	1.6	1.2	1.5	1.2
Other arterial disease	2.2	1.1	2.1	1
3. Respiratory diseases (CRD)
Bronchitis, Emphysema	12	11.8	17.1	15.6
Chronic airway obstruction	13.1	6.8	10.6	6.8
Pneumonia, Influenza	2.2	1.1	1.8	1.4
Diabetes mellitus	1.37	1.14	1.37	1.14
4. Tuberculosis (TB)*	1.57	1.57	1.57	1.57

Smoking-attributable deaths were calculated for each cause of mortality using the following formula;

SAM=OM ×PAF

Mortality attributable to tobacco smoking or SAM is the product of observed mortality (OM) and PAF. PAF is based on the following formula by Levin (1953):

PAF = {[p0 +p1RR1+p2RR2]-1} / [p0 +p1RR1+p2RR2]

Where p0, p1, and p2 are the prevalence rates of non-smokers, smokers, and ex-smokers respectively.

RR1 and RR2 are the risks of dying from any cause, for smokers and ex-smokers alike.

Morbidity (description of the analytical processes here)

This was a cross sectional study. The data on morbidity and economic costs was collected from 2,032 in and outpatients aged 18 years and above at four major national referral hospitals in Kenya – Kenyatta National Hospital (KNH), Moi Teaching and Referral Hospital (MTRH), Kenyatta National Hospital-Othaya Referral Hospital (KNH-ORH), and Kenyatta University Teaching, Referral, and Research Hospital (KUTRRH). All the interviewed patients suffered from one of the following 10 tobacco-related illnesses: cancer (oral-pharyngeal, laryngeal, lung), CVDs (myocardial infarction, stroke, peripheral arterial disease), COPD (chronic bronchitis, emphysema), diabetes, and pulmonary tuberculosis.

Sample Size and Sampling

We used the formula below to calculate the sample size:

n = Z2 p (1-p)/e2

Where n= sample size

Z= level of confidence

P= baseline level of the selected indicator (13% was the figure found in a study representing the proportion of tobacco use among tobacco-related disease admissions at KNH

)

e= margin of error

An adjustment for four age groups ( 35-54, 55-64, 65-74, and 75+ (1), number of facilities (4 facilities), and gender (2 categories) was made, therefore:e= margin of error

n2=n1*((4+4+2)-1) = 205*9 = 1,845

The sample size per facility was proportionately distributed according to their workload and bed capacity.

Distribution of sample size by hospital

Facility	Bed Capacity	Expected Workload	Sample Size
KNH	1,800	47%	868
MTRH	1,024	27%	494
KUTRRH	650	17%	314
KNH-ORH	350	9%	169
Total	3,824	100%	1,845

Distribution of sample size by hospital and disease condition

	Annual (2020) workload by disease condition (% workload)				Sample size allocation
Facility	Cancers	CVD	COPD	TB	Cancers	CVD	COPD	TB	Total
KNH	8,100 (78%)	1,144 (11%)	459 (4%)	673 (6%)	678	96	38	56	868
MTRH	3,790 (82%)	203 (4%)	299 (6%)	325 (7%)	406	22	32	35	494
KUTRRH	0 (82%)	0 (4%)	0 (6%)	0 (7%)	258	14	20	22	314
KNH-ORH	150 (12%)	600 (47%)	494 (39%)	23 (2%)	20	80	66	3	169
Total	12,040 (74%)	1,947 (12%)	1,252 (8%)	1,021 (6%)	1,366	221	142	116	1,845

Sampling procedure

This study used purposive sampling as it targeted patients with the selected TRI. Patient records, including outpatient and inpatient files and registers, were first checked to select patients with TRI. The inpatient and outpatient registers were reviewed daily in all the relevant wards and clinics during the study period. For outpatients, the study targeted all the adult outpatient specialty clinics, surgery and medicine wards, and all patients who met the criteria.

Data collection and analysis

Data collection was done by trained nurses/medical students in each hospital. Data was collected at the patient’s bedside in the wards, while outpatient study participants were interviewed at the different clinics where they were located. Data was collected using Kobo Toolbox (© Harvard Humanitarian Initiative) loaded in tablets. This data was collected offline and later uploaded to a central database. All electronic data was stored in password-protected databases and computers.

Statistical Package for Social Sciences (SPSS version 23.0) and Statistical Application Software (SAS version 9.0) were used to compute and analyze the study responses. Categorical characteristics were summarized as frequencies and percentages, with comparisons made between current smokers, occasional smokers, and never smokers and disease conditions with covariates using chi-square tests.

Economic costing

Study designs

This was a mixed study utilizing both quantitative and qualitative methods as well as key informant interviews (KII). The study was conducted in two phases.

Determination of cost

An activity-based costing approach was used to capture the costs of services relating to CVD, COPD, selected cancers, and pulmonary tuberculosis at the facility level. This approach measured all the resources, including healthcare workers, equipment, medications, and consumables, used to provide patients with required health services.

Steps in activity-based costing:

1. Selecting medical conditions under each of the four broad areas of CVDs, COPD, selected  cancers, and pulmonary tuberculosis.

2. Defining the care delivery value chain: charting out all the activities in the healthcare service delivery life cycle.

3. Creating process maps for each activity in the delivery of health services – recording every administrative and clinical process involved in the care of a particular medical condition.

4. Gauging the time spent and resources used in serving a patient at each point through the entire service chain (to arrive at time estimates for each process).

5. Estimating the expense of supplying patient care resources, which include direct and indirect care costs as well as overhead and support center costs.

6. Estimating the capacity of each resource and calculating the capacity cost rate and the total cost of patient care per disease.

Sample Size Determination

• Yamane’s formula, with maximum variability, 8% precision, and 95% confidence level, was used to determine the total sample size of 418 patients and clients.
• For the KII- three KIIs were done for each condition

Sampling

This study used a two-stage mixed sampling strategy combining purposive and random sampling methods. Employing purposive sampling, four referral hospitals were selected (as they represent the main hospitals providing services to the illnesses being costed). The second stage included random selection of patients with the selected TRI.

Data collection

The following data collection tools were used:

a) Key informant interview guide for CVDs, COPDs, selected cancers, and pulmonary tuberculosis

b) Process maps for adaption

c) Client exit interview survey

d) Facility cost data collection tool

The collection was done online via phones or tablets and using KOBO toolkit.

Data Analysis

The following steps were used to analyze the collected data:

Step 1: The service pathway map was first generated based on the collected facility data. The patient flow through different service delivery points was mapped.

Step 2: The inputs used to provide a service to a single patient or client were estimated and quantified. These inputs included the average time a patient spends with clinicians per facility visit. Other inputs included equipment, medicines, supplies, and tests.

Step 3: The capacity cost rate (CCR) of each resource (personnel, equipment, and space) was estimated. Data on the annual cost of each resource was obtained and divided by how often, in minutes, the resource could be used annually. Results were then summed to determine the unit cost per visit.

Step 4: Responses from the client exit survey were used to calculate the economic cost to the client. The economic cost included out-of-pocket health expenditures, direct transport costs, opportunity (indirect) costs for time spent at the facility, and transportation costs. The economic burden represented out-of-pocket expenditure and direct transport cost as a percentage of monthly household discretionary spending.

Step 5: The overall direct total cost of patient care was calculated by multiplying the CCR for each resource (personnel and space) by the average minutes that resource was used for each activity stage of care and the cost of supplies and equipment consumed at that stage. The cost per patient, or the unit cost per patient per service, was multiplied by the total number of patients to obtain the total cost per service.

Step 6: The above direct health costs were combined with the mortality and morbidity costs to estimate the economic cost of the diseases. Mortality and morbidity costs were estimated based on the productivity loss stemming from premature death and the inability to work due to morbidity.

Step 7: The direct costs (health costs) and indirect costs (costs related to lost productivity) were added to determine the total economic cost of each disease. The tobacco attributable factor was used to estimate the direct, indirect, and economic cost attributable to tobacco use.

Step 8:  To determine the net impact of tobacco use in the market, we estimated the net gain or loss by considering the tax collected by the government and the tobacco industry’s contribution to the economy. To obtain the aggregate revenue, we multiplied the retail price of cigarettes by aggregate consumption using secondary data. The tax, aggregate contribution (revenue), and the economic cost was compared to determine the net cost that tobacco imposes on society.

For the Agriculture, Industry Interference, Cessation and Enforcement themes, the TCDI team did not conduct data analysis but rather drew on secondary sources. The team undertook their initial data search and literature review between May 2021 and June 2022. Secondary data searches occurred across various platforms, including data repositories, academic journals, and online libraries. The team also reviewed resources from trusted organizations such as the Food and Agriculture Organisation, World Bank Group and government departments such as the Kenya National Bureau of Statistics. The TCDI team consulted with the Research Unit on the Economics of Excisable Products (REEP) and subject matter experts, who advised on data sources and reviewed each page before publication.