Describing and Modeling the Burden of Hospitalization of Patients With Neoplasms in Ghana Using Routine Health Data for 2012-2017.

PURPOSE: The increasing cancer burden calls for reliable data on current and future associated hospitalizations to enable health care resource planning, especially in low- and middle-income countries. We provide nationwide estimates of the current and future burden of hospitalization because of neoplasms in Ghana.
METHODS: We conducted secondary data (2012-2017) analysis using nationwide routine administrative inpatient health data from the Ghana Health Service. Multivariable Poisson regression was used to model spatial and temporal hospitalization trends stratified by sex and 5-year age group. In conjunction with official population projections, the model was used to predict future hospitalization up to 2032.
RESULTS: Out of 2,915,936 hospitalization records extracted for 6 years, 26,627 (1.0%) were for neoplasms, most of them benign (D10-D36, 15,362; 57.7%) and in female patients (20,159; 76%). In total, 9,463 (35.5%) patients with malignancies were mostly female (5,307; 56.1%), had a median age 50 years (interquartile range, 34-66 years) and a median duration of stay of 4 days (interquartile range, 2-8 days). Poisson regression for the malignant cancers revealed an annual increase in hospitalizations with a relative rate of 1.23 (95% CI, 1.19 to 1.27). The estimated hospitalization rate for malignancies of female patients was 1.5 times higher than that of male patients (relative rate, 1.53; 95% CI, 1.00 to 2.34), adjusted for age. We predicted an increase of 67.5% malignant cancer hospitalizations from the empirical years (2012-2017) into the prediction years (2022-2032) in Ghana.
CONCLUSION: In the absence of a national population-based cancer registry, this nationwide study used secondary health services data on hospitalizations as a proxy for neoplasm morbidity burden. Our results can support planning public health resources and building evidence-based advocacy campaigns for neoplasm-prevention efforts.

INTRODUCTION

The burden of noncommunicable diseases (NCDs) continues to show strongly disproportionate growth relative to population size irrespective of global efforts.[1-5] In 2018, the WHO estimated that 71% of the 40.5 million global deaths were due to NCDs, of which 34.4 million (85%) were premature deaths.[5] The majority of these premature deaths are estimated to have occurred in Africa where NCDs are associated with higher mortality than in high-income countries.[2,5,6] The disability burden of NCDs in sub-Saharan Africa from 1990 to 2017 showed a 67.0% increase in disability-adjusted life-years (DALYs),[6,7] from 90.6 million DALYs in 1990 to 151.3 million DALYs in 2017. It is estimated that by 2030, 46% of all deaths in Africa will be attributed to NCDs.[7,8] In Ghana, NCDs accounted for 43% of all deaths in 2016.[8,9] Cancers in particular show increased numbers globally as well in Ghana.[10] In 2020, Ghana recorded 24,009 incident cancer cases with corresponding mortality of more than two thirds (15,802) of the incident cases. Unfortunately, the GLOBOCAN estimates used only data from one of the two cancer registries in Ghana, which could have affected their estimates. Additionally, the reported mortality estimates were based on modeling of the mortality incidence from cancer registry data in neighboring countries.[10] Maintaining a cancer registry with nationwide coverage comes at a high cost to low- and middle-income countries (LMICs). An alternative, already existing data source to gain information on the cancer burden is the routine health care data on hospitalizations.

CONTEXT

Key Objective

A nationwide comprehensive analysis of neoplasm-related hospitalizations in Ghana using routinely collected health data from the District Health Information Management System 2.

Knowledge Generated

Overall, the hospitalization rate for females for any neoplasm was three times as high compared with male patients. The predicted future hospitalizations because of neoplasms is higher in females than in males for neoplasms with uncertain or unknown behaviors category. Hospitalization rates strongly increase with age in the 10 regions of Ghana for different types of neoplasms.

Relevance

Evidence-based advocacy campaigns targeted at the distribution of neoplasms across the different regions of Ghana and the identified sociodemographic risk factors should be initiated.

Because of the increasing burden of NCDs, it is imperative that health services plan and allocate resources accordingly—both toward treatment and prevention measures. However, this requires accurate and reliable estimates of the future burden of the disease. Routine health service delivery data are increasingly available but have been widely underused for research purposes, especially in LMICs.[11-13] The potential benefits gained from using these routinely collected health data sets for health services include morbidity and mortality estimation specific to geographic regions, planning of resource allocation, and acquisition and distribution of essential medical drugs, beds, and consumables. Administrative health data can also support research, aiding in public health and health care decisions and policy.[11,14,15] The Ghana Statistical Service data on vital statistics and population data, and the Ghana Health Service (GHS) District Health Information Management System 2 (DHIMS-2) database,[10] are two examples of databases that contain nationwide information on vital statistics and patient health records.

Cancer continues to be the second leading cause of NCD morbidity and mortality globally, accounting for 9.6 million (22%) annual deaths with about 6.7 million (70%) of these deaths in LMICs.[16] Cancer is estimated to account for 4,720 (5%) annual deaths in Ghana.[8,9] Several small and nonrepresentative hospital-based studies have indicated that cancer of the breast, prostate, and cervix[10,17-22] are among the leading neoplasms responsible for hospital admissions. However, reliable nationwide data in Ghana are missing such that there is large uncertainty about the future cancer-related morbidity, mortality, and hospitalizations. Similar to previous data sets used for estimating NCD burden,[1,2,10] the DHIMS-2 data set is a promising candidate, but has not yet been applied in detail to the situation in Ghana. Hence, our study aimed to provide reliable nationwide age, sex, and region-specific estimates of the current and future burden of hospitalization because of neoplasms in Ghana.

METHODS

Study Setting and Data Sources

This study was based on data from two sources. First, inpatient data on neoplasms were extracted from the GHS DHIMS-2 database. The DHIMS-2 is based on the District Health Information System 2 developed by the University of Oslo, Norway, which is widely used in LMICs.[23] It covers inpatient data entered by more than 10,000 users from over 300 university/teaching, regional, district, and mission/private hospitals in Ghana.[24] Clinicians or data entry staff are supposed to code patient diagnoses in the DHIMS-2 according to the WHO International Statistical Classification of Diseases and Related Health Problems (International Classification of Diseases, 10th Revision [ICD-10]) classification.[25] However, this is not always the case as some of the diagnoses are captured as free text without ICD-10 codes. The DHIMS-2 is an improvement of DHIMS-1, which lacks the ICD-10 codes and is not user friendly. In Ghana, neoplasm-related diagnoses vary considerably across the different types of hospitals. For instance, benign and malignant excisions are done by medical officers in district hospitals and sent for histopathology in another hospital or laboratory with the capacity to do the test. Although benign tumors are managed in local hospitals, malignant tumors generally are referred to larger and teaching hospitals. In regional hospitals, which serve as the referral points for district hospitals, neoplasms requiring excisions are treated by surgeons, with histopathology and management either done within the hospital or referred to a specialized center. Ghana does not have national cancer registry, and the two population-based registries are either not well defined or limited in scope like in most LMICs.[10,20,26-28] The DHIMS-2 is one of the few nationwide data sources for disease-specific hospitalizations globally. Hence, for our study, 13 variables on hospital admission and discharge including sociodemographic characteristics were extracted by the national data manager with assistance from the principal investigator. Second, the Ghana Statistical Service provided past and future projected counts for the country's population by region, age, and sex.[29]

Study Population

The study participants included patients hospitalized with all types of neoplasms from 446 hospitals between 2012 and 2017 recorded in the DHIMS-2 database. Patients with missing data on age, sex, or date of admission were excluded as these could be described as missing at random.

Statistical Analysis

All neoplasms were defined by the principal diagnosis for hospitalization according to ICD-10 neoplasm codes (C00-D48). Missing ICD-10 codes captured as free text (approximately 10%) were manually assigned on the basis of the free text, where possible. Hospitalization duration was defined as the difference between the date of admission and discharge plus one. Each episode of hospitalization was treated as a unique record as our data did not distinguish between multiple visits per patient. The sociodemographic characteristics of the patients were stratified by the malignancy category, that is, malignant (C00-C97), benign (D10-D36), and neoplasms of uncertain or unknown behaviors (D37-D47). We modeled the rate of hospitalizations because of neoplasms between 2012 and 2017 using Poisson regression with a robust sandwich estimator to account for overdispersion. Our choice of Poisson regression is to report (empirical) and predict (future) rates of neoplasm hospitalizations as a count data compared with other models.[30,31] Modeling the future hospitalization rates is useful to hospital managers and policymakers for resource allocation. The three malignancy groups (malignant, benign, and neoplasms of uncertain or unknown behaviors) were treated as categorical outcomes in three separate models. Covariates in the model included calendar-year (quantitative, reference year 2012), age (categorized into 5-year age groups with age 25-29 years as the reference group), sex (reference group: male), region (10 categories with the Greater Accra region as the reference group), as well as an interaction term for age and sex. Log population count was used as the offset. This model was used to predict the future (2022 through 2032) hospitalization counts for three broad groups of neoplasms on the basis of the projected population data.

For the predictions, we assumed no effect of calendar year beyond 2016. This assumption implies that any previous secular trend in health care delivery or diagnostic quality associated with neoplasm-related hospitalization rates reached a plateau in 2016. The assumption also implied that potential risk factors for neoplasm incidence not included in the regression model would remain constant. Hence, the only variables with accommodated future changes associated with neoplasm hospitalization rate were projected population counts per region, age, and sex. We assessed how well the model fits our data using visual inspection of the Pearson residual plot for each group of neoplasms.

We report the empirical (2012-2017) and predicted (2022-2032) hospitalization counts with 95% CIs and corresponding P values as well as relative rates (RRs) with 95% CI and P values from Wald tests.

Data cleaning, analysis, and predictions were performed using Stata v15.[32] Stata codes, as well as the GHS patient and Ghana Statistical Service population data sets, can be obtained from the principal investigator upon request.

The study was approved by the GHS Ethics Review Committee (GHS-ERC: 002/12/17).

RESULTS

Participants

A total of 2,915,936 inpatient records were identified from 2012 to 2017. After excluding all cases for other diagnoses (n = 2,887,164), records with incomplete data (n = 1,383‬), and duplicate records (n = 762), the study population comprised 26,627 records of hospitalization for neoplasms (Fig 1). The overall sociodemographic characteristics of patients are presented in Table 1, and detailed breakdowns of all neoplasms are shown in the Appendix Table A1.

FIG 1

Study flowchart of hospitalized patients in the Ghana Health Service DHIMS-2 database, 2012-2017. DHIMS-2, District Health Information Management System 2.

TABLE 1

Characteristics of Patients Hospitalized for Cancer Between 2012 and 2017 in Ghana

TABLE A1

Breakdown of Neoplasms by Broad Categories

Description of Patients Hospitalized for Neoplasms

From Table 1, out of the 26,627 hospitalization records analyzed in this study, 9,463 (35.5%) were due to malignant disease, 15,362 (57.7%) due to benign neoplasms, and 1,802 (6.8%) due to neoplasms of unknown or uncertain behaviors (D37-D47). The median age among patients with malignancies was 50 years (interquartile range [IQR], 34-66 years), among patients with benign neoplasms was 39 years (IQR, 32-45 years), and among patients with neoplasms of uncertain or unknown behaviors was 37 years (IQR, 22-55 years). The median duration of hospitalization was 4 days irrespective of malignancy status. However, the IQR for patients with malignancies ranged from 2 to 8 days and that of patients with benign disease ranged from 2 to 5. The median duration of stay was 3 days for patients with neoplasms of unknown or uncertain behaviors. For patients with malignant and benign diseases, the annual number of hospitalizations increased from 714 (7.6%) and 134 (0.9%) in 2012 to 2,274 (24.0%) and 4,998 (32.5%) in 2016, respectively. Although there was an increase in hospitalization because of neoplasms of unknown or uncertain behaviors (D37-D47) from 2012 to 2013, it kept fluctuating for the subsequent years.

The regional hospitalization rates showed that the Volta region recorded the highest proportions of admissions for both malignant (1,652; 17.5%) and benign disease (3,349; 21.8%) admissions. Among patients with neoplasms of unknown or uncertain behaviors, most of the records were from the Greater Accra region (380; 21.1%). Most of the patients with any of the neoplasms either had no education (6,437; 24.2%) or completed primary education (11,656; 43.8%). Unemployment and health insurance ownership were reported by the majority of patients (15,128 [56.8%] and 23,139 [86.9%]). Comorbidity (2,476; 26.2%) and case fatality (1,016; 10.7%) were highest among patients with malignancies. The lowest comorbidity proportion was among patients with neoplasms of uncertain or unknown behaviors (283; 15.7%), and the lowest case fatality among patients with benign disease (60; 0.4%). Since this was a nationwide study, hospitals classified as district/government/municipal (11,290; 42.4%) or regional (6,739; 25.3%) accounted for most hospitalizations, especially for those because of benign neoplasms. The six university/teaching hospitals included in the study recorded more neoplasms of uncertain or unknown behaviors (109; 6.1%), malignant (434; 4.6%), and benign (433; 2.8%). The remaining sections of the results focus on malignant neoplasms as the benign tumors could be described as mainly precancerous lesions.

Modeling Past Hospitalizations (2012-2017)

The Poisson regression for hospitalization because of any neoplasm showed an annual RR of 1.43 (95% CI, 1.36 to 1.52; P < .001). The overall rate of female patients was three times compared with that of male patients (RR, 3.04; 95% CI, 2.11 to 4.37; P < .001), modified by age. In the age-sex-specific strata, females in age groups 20-24 years to 45-49 years have higher RRs between 1.29 and 2.29 compared with males. In the other age groups, females have lower age-sex-specific strata RRs between 0.15 and 0.94 compared with males. The remaining overall results are shown in Figure 2 and Appendix Table A2.

FIG 2

Coefficient and 95% CI plot from the Poisson regression model for patients hospitalized because of neoplasms: (A) neoplasms, (B) C00-C97, (C) D10-D36, and (D) D37-D47.

TABLE A2

Relative Rates, 95% CI, and P Value From the Poisson Regression Model for Neoplastic Patients Hospitalized Between 2012 and 2017

The annual RR for hospitalization because of malignant neoplasms (C00-C97) was 1.23 (95% CI, 1.19 to 1.27; P < .001). For malignant neoplasms, the RR for female patients compared with male patients was 1.53 (95% CI, 1.00 to 2.34; P = .048).

Compared with patients age 25-29 years, diagnosed with malignant, younger patients had lower hospitalization rates and older patients had higher hospitalization rates. Among the younger patients with malignant diagnoses, the RRs ranged from 0 to 4 years (RR, 0.37; 95% CI, 0.23 to 0.59; P < .001) to 45-49 years (RR, 2.28; 95% CI, 1.51 to 3.44; P < .001), whereas much higher rates were observed among older patients in the range of (RR, 2.89; 95% CI, 1.91 to 4.39; P < .001) from 55 to 59 years to (RR, 22.84; 95% CI, 15.04 to 34.70; P < .001) among patients age 75-59 years. The RRs in each neoplasm category increased with age, especially among the high-risk group compared with patients age 25-29 years (Fig 2 and Appendix Table A2).

All regions except the Eastern Region showed a statistically significant RR compared with the Greater Accra region (reference region). The highest RR (1.33; 95% CI, 1.09 to 1.61; P = .004) and lowest RR (0.24; 95% CI, 0.18 to 0.32; P < .001) for hospitalization because of malignant neoplasms were in the Volta and Upper West regions, respectively.

The RRs for age and sex interaction were significantly lower for females < 10 years and those above 60 years with malignancies. Rates for the other neoplasms are shown in Figure 2 and Appendix Table A2.

Predicting Future Neoplastic Hospitalizations (2022-2032)

We present the predictions of malignant neoplasms (C00-C97) here, whereas the predictions of benign and neoplasm of uncertain or unknown behaviors are shown in Appendix Figures A1 and A2.

FIG A1

Three main blocks of neoplasms according to the International Classification of Diseases, 10th Revision, classification of neoplasms in the Ghana Health Service District Health Information Management System 2 database, 2012-2017: (A) male and (B) female.

FIG A2

Yearly number and 95% CI of empirical and predicted hospitalizations because of neoplasms of uncertain or unknown behaviors from Poisson regression model by sex: (A) male and (B) female.

The visual inspection of the Pearson residual plot for each group of neoplasms showed no variation between the observed and the fitted values according to Appendix Figures A3A-A3C.

FIG A3

Pearson residual plot for (A) malignant cancers, (B) benign tumors, and (C) neoplasms of uncertain or unknown behaviors.

From 2012 to 2016, the annual number of hospitalizations among males were 442 (319 malignant) to 1,503 (1,002 malignant), and among females 587 (415 malignant) to 6,020 (1,272 malignant). We predict that the yearly number of hospitalizations of male patients with malignant neoplasm in 2022 will be 1,349 (95% CI, 1,285 to 1,413), corresponding to a 34.6% increase compared with 2016. For females, we predict 1,776 (95% CI, 1,715 to 1,837) hospitalizations, representing a 39.6% increase compared with 2016. For 2026, the projected number of hospitalizations of male patients with malignant neoplasm is 1,470 (95% CI, 1,391 to 1,549) corresponding to a 46.7% increase compared with 2016. For females, the prediction is 1,926 (95% CI, 1,849 to 2,002), representing a 51.4% increase compared with 2016. In 2032, the corresponding predictions are 1,655 (95% CI, 1,549 to 1,760) for males and 2,155 (95% CI, 2,052 to 2,258) for females (Fig 3 and Table 2), representing a 65.2% and 69.4% increase in males and females, respectively, compared with 2016.

FIG 3

Yearly number and 95% CI of empirical and predicted neoplastic hospitalizations from Poisson regression model by sex: (A) male and (B) female.

TABLE 2

Yearly Number and 95% CI of Empirical and Predicted Cancer Hospitalization Cases From Poisson Regression Model by Sex

DISCUSSION

We demonstrated the usefulness of routinely collected health data (2012-2017) in estimating the current and future burden because of hospitalizations for neoplasms in Ghana in the absence of a national population-based cancer registry.[6,18,28] To our knowledge, our study is the first to provide a nationwide estimate of hospitalizations for all different neoplasms among different hospital types in Ghana. The documented annual number of all hospitalizations of patients with neoplasms increased by 672.8% in 6 years. The increase was largely influenced by benign tumors. This large increase could partly be explained by the ambitious efforts by the GHS to improve its health information system in 2012.[33] It should be noted that only hospitalization rates are available at the national level as no national cancer registry currently exists. Additionally, cancers and other NCDs in sub-Saharan Africa are estimated to increase.[6]

The burden because of hospitalization of women with neoplasms was three times higher than that of men, consistent with the results from smaller studies on the demographics of patients with cancer.[17,18,21] However, the rates for hospitalization of female patients with malignant disease were about twice that for men, whereas hospitalization rates of women with benign disease were six times those of their male counterparts. The higher proportion of female patients with neoplasms was obvious across the 10 regions of Ghana, which could probably be explained by high cervical and breast neoplasms. This distribution is also consistent across the different hospital types involved in the study. Patients with malignancies were a decade older than the other patients in our study, in contrast to previous smaller studies in Ghana.[17,18,26] On the basis of regional population distribution, one would expect that the highest number of all neoplasms would have been recorded in the Greater Accra and Ashanti regions. However, this was not the case. The highest number of hospitalizations for both malignant and benign neoplasms was in the Volta region. Presumably, this could be due to a specialized hospital with foreign assistance in caring for patients with fibroid, making the hospital accessible to the whole country, especially to the people of the Greater Accra region and the Republic of Togo. Also, a probable reason for the low number of hospitalizations in the two biggest regions (Greater Accra and Ashanti) could be due to the teaching hospitals with associated cancer registries in these two regions,[17,28] which do not contribute data to the DHIMS-2.

The overall high increase in the observed and predicted burden of the different neoplasms could be plausibly because of the high burden of leiomyoma of uterus (unspecified), and female malignant neoplasm of the breast, cervix uteri, prostate, and ovary compared with the other neoplasms.[16-18,21,26,28,34]

The Poisson regression model showed a 23% (RR, 1.23) increase in the annual rates for malignant neoplasm admissions over the 6 years. The Volta region recorded the highest RRs of 1.33 for patients with malignancies. The stronger increase in hospitalization rates of patients between ages 30 and 75+ years for malignant neoplasm and neoplasms of uncertain or unknown behaviors reflects the general age distribution of cancer incidence previously reported from some teaching hospitals in Ghana.[17,18]

We predicted the number of neoplastic-related hospitalizations over 10 years until 2032 on the basis of the available population forecast for this time period. Nevertheless, limiting the prediction to a shorter time interval is preferred because of large uncertainties in very long-term population estimates.[35-37] In our predictions, we estimated that 1,349 and 1,776 male and female patients with malignant cancers would be hospitalized in 2022. Subsequently, the year-to-year relative increase in the number of predicted hospitalization counts increases from 2.1% in 2023 to 6.4% in 2026 and doubles in 2032 for patients with malignancies. Hospitalizations of male patients were predicted to slightly increase by 22.7%, compared with 21.3% for females. These estimates were made assuming that the same level of health care delivery in 2016 is maintained as is (doctor-to-patient ratio, health education, and availability of health facilities), and lifestyle risk factors remain unchanged. The year 2016 was chosen as the reference year, as the DHIMS-2 database had incomplete data for 2017. Assuming constant external factors on hospitalization because of neoplasms, leaving only demographic development driving the change in hospitalization rates from 2016 onward represents a conservative compromise, given the large uncertainty in making specific predictions about the relevant individual influences. It aims to balance a possibly increasing trend in health care availability with a possible success of prevention efforts reducing all neoplasm incidence.[35,37-39] We also did not provide predictions from years 2017 through 2021 because this period elapsed while our study was ongoing.

In line with similar global estimates,[1,5,7] we extended our predictions to 10 years, although this increases the uncertainty.[39-41] Our long-term predictions aim to raise awareness of the expected increase in the burden of all neoplasms among inpatients across the different regions and corresponding challenges to health services planning. Our predictions will also allow for future comparison studies with observed data to benchmark and validate attempts to use administrative health data for future resource planning.

The overall increase in the estimated burden of the different neoplasms among hospitalized patients is consistent with the globally estimated cancer burden.[2,5,7]

It should be noted that we were not able to calculate incidence rates and that hospitalization may only be partly correlated with incidence. A major limitation of this study is the fact that most of the neoplasms especially the malignancies were diagnosed as unspecified. In addition, the two biggest teaching hospitals in Ghana do not contribute data on neoplasms to the DHIMS-2 database. However, the majority of the inpatients from these two hospitals also receive medical care in the majority of hospitals involved in this study before patient referral. Also, about 70% of the benign cases in the teaching hospitals are referred from district and regional hospitals, minimizing the effect further. However, the effect on our estimates with regards to hospitalization because of malignant neoplasms may be moderate. Further limitations are implied by the unavailability of modifiable risk factors in the DHIMS-2 database for our study population. This could bias the prediction in the case of a drastic change in these factors. Another important fact to consider is the dominance of fibroid, breast, cervical, and prostate cancers, which, when detected early,[16] can reduce the overall number of hospitalizations.

The interpretation of our predictions must be done with caution, as future changes in the diagnostic, therapeutic, and preventive conditions could result in higher or lower hospitalization rates. Our prediction of neoplasm-related hospitalization rates in Ghana is an essential support to the public health measures, planning hospitals' capacity to admit the increasing number of neoplastic patients, and to estimate the national burden of all neoplasms among hospitalized patients.[37,42]

In conclusion, the reported study has three major implications: resource allocation, improved data on all neoplasm-related hospitalizations and neoplasm-related policies in Ghana, and usefulness of routine health databases. First, the projected nationwide hospitalization trend for malignant cancers, benign, and neoplasms of uncertain or unknown behavior could be used in efficient health care planning, resource allocation, and hospital management. Second, the data from this study are a strong contribution to the body of knowledge on the types of neoplasms in Ghana and the usefulness of the DHIMS-2 database. Finally, the empirical and estimated hospitalizations because of neoplasms could possibly be used in prioritizing neoplasm-related policies such as early detection measures, prevention programs, and outpatient management.

Although not all private hospitals admitting patients with neoplasms contribute data to the GHS database, we believe that all the regional, district, mission, and the private hospitals on the DHIMS-2 platform are adequate for us to generalize our findings to cover the whole country. The evidence from our study also paves the way for future nationwide studies assessing the diagnostic, treatment, and management capacities of the various hospitals other than the teaching hospitals to which neoplastic patients seek care.